Passive FTIR Phase I Testing of Simulated and …...Passive FTIR Phase I Testing of Simulated and Controlled Flare Systems FINAL REPORT Prepared for: Texas Commission on Environmental

Passive FTIR Phase I Testing of

Simulated and Controlled Flare Systems

FINAL REPORT

Prepared for:

Texas Commission on Environmental Quality 12100 Park 35 Circle

Austin, TX 78753

University of Houston 4800 Calhoun Road

Houston, TX 77204-2011

Prepared by:

URS Corporation 9400 Amberglen Blvd.

Austin, TX 78729

URS Corporation 9801 Westheimer, Suite 500

Houston, TX 77042

June 2004

. ii

Table of Contents 1.0 Executive Summary ......................................................................................................... 1-1 1.1 Rationale for Selecting Passive FTIR.................................................................. 1-1 1.2 Program Elements of Phase I Study..................................................................... 1-2 1.3 Data Quality Objectives....................................................................................... 1-3 1.4 Principles of Passive Remote Sensing Using PFTIR........................................... 1-4 1.5 Evaluating Simulated Flare Emissions from Plume Generator ........................... 1-5 1.6 Summary of Key Observations and Conclusions ................................................ 1-7 1.6.1 Analytical Study Findings........................................................................ 1-7 1.6.2 Phase I Plume Generator Validation Study Findings .............................. 1-9 1.6.3 Controlled Flare Experiment Findings................................................... 1-11 1.7 Path Forward Recommendations ....................................................................... 1-12 2.0 Introduction...................................................................................................................... 2-1 2.1 Background.......................................................................................................... 2-1 2.2 Objectives ............................................................................................................ 2-3 2.3 Organization of Report ........................................................................................ 2-3 3.0 Test Facility Description and Process Operation............................................................. 3-1 3.1 Plume Generator System...................................................................................... 3-1 3.2 Controlled Flare System ...................................................................................... 3-3 3.3 Process Operation During Testing ...................................................................... 3-7 3.3.1 Plume Generator ...................................................................................... 3-7 3.3.2 Controlled Flare ....................................................................................... 3-9 4.0 Sampling and Analytical Procedures ............................................................................... 4-1 4.1 Sampling Procedures ........................................................................................... 4-1 4.1.1 Airflow Metering and Gas Spiking – Plume Generator Tests ................. 4-1 4.1.2 Extractive FTIR (EPA Method 320) - Plume Generator Tests................ 4-2 4.1.3 Canisters................................................................................................... 4-2 4.1.3.1 Plume Generator Tests.............................................................. 4-5 4.1.3.2 Controlled Flare Tests............................................................... 4-5 4.1.4 Passive FTIR............................................................................................ 4-6 4.1.4.1 Plume Generator Tests.............................................................. 4-6 4.1.4.2 Controlled Flare Tests............................................................. 4-12 4.2 Analytical Procedures ........................................................................................ 4-14 4.2.1 Extractive FTIR - Plume Generator Tests ............................................. 4-14 4.2.1.1 Calibration and Quality Control Checks................................. 4-18 4.2.2 Canisters................................................................................................. 4-22 4.2.2.1 Plume Generator Tests............................................................ 4-22 4.2.2.2 Controlled Flare Tests............................................................. 4-22 4.2.3 Passive FTIR.......................................................................................... 4-23 4.2.3.1 Principles of Passive FTIR Measurement............................... 4-23 4.2.3.2 Data Analysis Techniques ...................................................... 4-28

. iii

Table of Contents (continued) 4.2.3.3 PFTIR Calibration................................................................... 4-35 4.2.3.4 PFTIR Validation and Quality Control................................... 4-42 5.0 Presentation and Discussion of Results ........................................................................... 5-1 5.1 PFTIR Analytical Study....................................................................................... 5-1 5.1.1 Radiant Emission Signal Levels .............................................................. 5-1 5.1.2 Minimum Detection Limits...................................................................... 5-2 5.1.3 Plume Gradients....................................................................................... 5-7 5.2 Plume Generator Testing ..................................................................................... 5-9 5.2.1 Stratification Test................................................................................... 5-11 5.2.2 Emissions Tests...................................................................................... 5-15 5.2.2.1 Concentration Measurement Data........................................... 5-15 5.2.2.2 Combustion Efficiency Results............................................... 5-26 5.2.3 Precision Test......................................................................................... 5-29 5.3 Controlled Flare Testing .................................................................................... 5-29 5.3.1 Flare Fuel Analysis ................................................................................ 5-30 5.3.2 Flare Passive FTIR Measurements ........................................................ 5-31 5.4 Measurement Uncertainty and Quality Control................................................. 5-33

5.4.1 Measurement Uncertainty ..................................................................... 5-34 5.4.1.1 EFTIR Analytical Uncertainty................................................ 5-34 5.4.1.2 PFTIR Uncertainty.................................................................. 5-35

5.4.2 Quality Control ...................................................................................... 5-36 5.4.2.1 Canister Samples..................................................................... 5-36 5.4.2.2 EFTIR Measurements ............................................................. 5-39 5.4.2.3 PFTIR Measurements ............................................................. 5-44 5.4.2.4 Process Instrumentation .......................................................... 5-46 5.5 Statistical Analysis of Plume Generator Test Results........................................ 5-47 5.5.1 Analytical Uncertainty of the Combustion Efficiency Results.............. 5-48 5.5.2 Test for Bias in the PFTIR Method........................................................ 5-49 6.0 Summary, Conclusions, and Recommendations.............................................................. 6-1 6.1 Summary .............................................................................................................. 6-1

6.2 Conclusions.......................................................................................................... 6-3 6.3 Recommendations................................................................................................ 6-7

. iv

Appendices Appendix A QAPP

Appendix B Extractive FTIR Data

Appendix C Passive FTIR Data and Calculations

Appendix D John Zink Process Data

Appendix E Sampling Equipment Calibration Data

Appendix F Canister Analyses

. v

List of Tables 1-1 Target Test Conditions for Plume Generator Experiments.............................................. 1-7 3-1 Plume Generator and Spiking System Process Settings .................................................. 3-8 3-2 Statistical Summary of Process and Meteorological Data for Controlled Flare Test (5:12 PM to 6:03 PM)...................................................................................................... 3-9 4-1 Target Test Conditions for the Plume Generator Tests ................................................... 4-3 4-2 Canister Samples Collected During the Plume Generator Tests ..................................... 4-5 4-3 Canister Samples Collected During the Plume Generator Tests ..................................... 4-6 4-4 Positions Monitored by the PFTIR in the Flare Plume, Averaging Times, and Average Temperature Observed ............................................................................. 4-13 4-5 EFTIR Analytical Method Parameters for Target Compounds and Spectroscopic Interferants ..................................................................................................................... 4-17 4-6 EFTIR MDLs for Target Compounds............................................................................ 4-18 4-7 EFTIR Calibration Gas Standards ................................................................................. 4-19 4-8 PFTIR Conformance to TCEQ Quality Control Requirements..................................... 4-43 5-1 Radiance Levels Resulting from the Plume-Radiance Simulations ................................ 5-2 5-2 Minimum Detectable Gas Concentrations from Simulations at 150°C, 225°C, and 232°C ............................................................................................................................... 5-6 5-3 Analysis of Spectra Generated Using the Three Profiles Shown in Figure 5-3............... 5-9 5-4 Test Schedule for the Plume Generator Tests................................................................ 5-11 5-5 EFTIR Stratification Test Data ...................................................................................... 5-13 5-6 Plume Generator Tests - Spiking Gas Concentrations by Calculation and EFTIR

Measurement.................................................................................................................. 5-16 5-7 Plume Generator Tests - Comparison of Average Gas Concentrations........................ 5-18 5-8 Plume Generator Test 3a................................................................................................ 5-20 5-9 Simulated Combustion Efficiency Comparison Between EFTIR and PFTIR Measurement Calculations............................................................................................. 5-27 5-10 PFTIR Results for Test 3b Replicate Measurements..................................................... 5-29 5-11 Measurement Activities for the Controlled Flare Test (August 27) .............................. 5-30 5-12 Canister Sample Results for Propane Fuel..................................................................... 5-31 5-13 Controlled Flare Test PFTIR Data Summary ................................................................ 5-32 5-14 Uncertainties for Analytes of Interest ............................................................................ 5-35 5-15 Comparison of PFTIR Measurement Uncertainties to Field Blanks ............................. 5-36 5-16 QC Results for Plume Generator Canister Samples ...................................................... 5-37 5-17 QC Results for Propane Fuel Canister Samples ............................................................ 5-38 5-18 EFTIR Method 320 Spiking Results for Certified Gas Standards................................. 5-41 5-19 EFTIR Heated Air Background Measurements............................................................. 5-42 5-20 Analytical Uncertainties for Combustion Efficiencies .................................................. 5-49 5-21 Pairwise Differences of Combustion Efficiency Measurements ................................... 5-51 5-22 Summary of Plume Generator Test Results................................................................... 5-53

. vi

List of Figures 1-1 Photograph of Plume Generator used to Evaluate PFTIR ............................................... 1-6 1-2 Photograph of Passive Fourier Transform Infrared (PFTIR) Spectrometer .................... 1-8 3-1 Schematic Diagram for the Plume Generator System ..................................................... 3-4 3-2 Schematic Diagram for the Plume Generator System (Propane, CO and CO2 injection test) .............................................................................. 3-5 3-3 Schematic Diagram for the Controlled Flare System ...................................................... 3-6 3-4 Controlled Flare Process and Meteorological Data During the 8/27/03 Test Period .... 3-10 4-1 Schematic of Stack Gas Extractive FTIR System............................................................ 4-4 4-2 Photograph of Passive Fourier Transform Infrared (PFTIR) Spectrometer .................... 4-7 4-3 Photograph of Plume Generator Used to Evaluate PFTIR .............................................. 4-9 4-4 Angle of PFTIR Line of Sight at the Stack.................................................................... 4-10 4-5 Relative Position of FPTIR Field of View Across the Top of the Stack ....................... 4-10 4-6 Observed Signal Profiles Across the Simulator Stack................................................... 4-11 4-7 Schematic Diagram of EFTIR Apparatus Measurement ............................................... 4-16 4-8 Planck Function Showing the Radiation Emitted by a Blackbody at 200°C................. 4-25 4-9 Contributions to Flare Radiance .................................................................................... 4-26 4-10 Structure of the Fundamental CO Band at 300K (top) and 500K (bottom) Showing Alteration of Band with Temperature............................................................. 4-29 4-11 Measured FTIR Signal Using the Black Body Source at 150°C, 250°C....................... 4-36 4-12 Background Fits to the Spectra of Figure 4-11 Eliminating at Atmospheric Absorption Features ....................................................................................................... 4-36 4-13 System Response Determined form 150°C, 250°C and 316°C Measurements............. 4-37 4-14 Spectral Radiance Obtained on the Flare Test at 17:14:26 on 27 August 2003 ............ 4-40 4-15 Flare Test Radiance Data and Radiance Fit in the CO2 and CO Emission Spectral Regions ............................................................................................................ 4-40 4-16 Flare Test Radiance Data and Radiance Fit in the Ethylene and Propylene Emission Spectral Region .............................................................................................. 4-41 4-17 Flare Test Radiance Data and Radiance Fit in the Butane Emission Spectral Region .. 4-41 5-1 Synthetic Radiance Spectra for a Low Efficiency, High Temperature Flare Simulation and a High Efficiency, Low Temperature Flare Simulation ......................... 5-3 5-2 PFTIR System Response in (watts/cm2/str./cm-1) per Volt of FTIR Output ................... 5-4 5-3 Three Plume Temperature Gradients Modeled during the Analytical Study .................. 5-8 5-4 Calculated CO2 and CO Radiance Levels at 498K for the Three Profiles of Figure 5-3......................................................................................................................... 5-8 5-5 Log Plot of CO Rotational-line Energy Versus Line Quantum Number....................... 5-10 5-6 Stratification Tests Results ............................................................................................ 5-14 5-7 Time Series CO2............................................................................................................ 5-21 5-8 Time Series for CO ........................................................................................................ 5-22 5-9 Time Series for Butane .................................................................................................. 5-23 5-10 Time Series for Ethylene ............................................................................................... 5-24 5-11 Time Series for Propylene ............................................................................................. 5-25 5-12 Calculated Combustion Efficiencies with Uncertainty Bars for EFTIR and PFTIR..... 5-28 5-13 Efficiency Measurements for Test Condition 3a ........................................................... 5-51 5-14 Distribution of Paired Efficiency Differences for Test 3a ............................................. 5-52

1-1

1.0 Executive Summary

The Texas Commission on Environmental Quality (TCEQ) is working to improve emission estimates for flares. Currently, these estimates are based on emission factors that were derived from limited data obtained as part of EPA-sponsored testing in 1980. Flare emissions may vary based on actual flare operation, and there may be more variables that affect flare operation than were identified in the 1980 studies. Thus, it is desirable to be able to determine speciated emissions and combustion efficiency during actual operation. Therefore, the TCEQ has contracted the University of Houston (UH) and URS Corporation (URS) to evaluate the feasibility of Passive Fourier Transform Infrared (PFTIR) spectroscopy as a candidate method for measuring flare emissions, and to use those measurements to calculate combustion efficiency of flares based on the following equation:

[soot][THC][CO]][CO

][COEff2

2

+++=

where: [CO2] is the CO2 concentration, [CO] the CO concentration, [THC] the concentration of total hydrocarbons, and [soot] is the concentration of any soot present. In cooperation with TCEQ and UH, the URS teamed with Industrial Monitor and Control

Corporation (IMACC) and John Zink Company (John Zink). The team developed a multi-phase study for the purpose of assessing if the PTFIR is a viable candidate method for determining the combustion efficiency and total speciated mass emissions from operating process flares with a known level of accuracy and precision. The first phase of the study was to evaluate the ability of the PFTIR to measure concentrations of simulated flare emissions in a controlled test environment. The second phase will be to evaluate the PFTIR technology on actual process flares operating in the HGA. This report summarizes the results of the Phase I testing.

1.1 Rationale for Selecting Passive FTIR Traditional extractive sampling methods generally collect an aliquot of the pollutant gases or species of interest from within a well-mixed exhaust stack prior to their release to the atmosphere. In most cases these exhaust stacks are equipped with platforms and sampling ports to permit easy access for the sampling equipment and personnel. This permits a variety of continuous or integrated measurement techniques to be used to quantify the emissions from these

1-2

sources. Because the combustion of industrial flares occurs at the flare tip and the exhaust gases are emitted directly to the atmosphere at a height of several hundred feet; use of traditional stack sampling methods for characterizing flare emissions are not practical. Passive Fourier Transform Infrared (PFTIR) spectroscopy was selected as a candidate method for sampling flares for three reasons. First, passive remote sensing using PFTIR offers the possibility of characterizing flare emissions non-intrusively and at a distance. This approach eliminates the need for special cones, sampling rakes, and lifting devices to hoist the sampling packages into position over the flare plume. All of these are manpower intensive and logistically complicated. Secondly, PFTIR may be capable of cost effectively quantifying major constituents of the flare emissions in industrial settings. As compared with other types of remote sensing devices, the PFTIR can quantify many compounds simultaneously (many of which are products of complete and incomplete combustion), thus the measurements can be made more cost-effectively. Finally, the PFTIR approach may provide a method for directly assessing flare performance continuously and in near real-time. This could be very advantageous when measuring flares that may be over steamed (or air assisted), or when characterizing the effects of wind speed on flare efficiency. 1.2 Program Elements of Phase I Study The Phase I test program was comprised of three major elements. These included:

• An analytical study to evaluate the performance characteristics of the PFTIR method for conducting flare emission measurements in the field;

• A controlled field test to determine the accuracy and precision with which PFTIR can measure known emissions from a simulated flare under actual field conditions; and

• The use of PFTIR to measure emissions from a controlled industrial flare to evaluate the logistical difficulties in making actual flare measurements in the field.

Prior to deploying to the field an analytical study was conducted to assess the performance characteristics anticipated for a PFTIR system. This study included:

• Determining the expected signal levels for the various species of interest as a function of gas temperature and gas concentration;

• Determining the minimum detectable concentrations possible for C2-C4 compounds, carbon monoxide, carbon dioxide, and C5 + hydrocarbons at typical flare operating conditions; and,

1-3

• Evaluating the maximum combustion efficiency that can be calculated using PFTIR at various gas temperatures.

The analytical study served as a useful first step in assessing the expected performance of

the PFTIR to meet the overall program objectives, and provided valuable information needed to properly configure the PFTIR during the Phase I field study. The second element of the study was to evaluate the performance of the PFTIR to measure simulated flare plumes (while varying the concentration of selected compounds and stack gas temperature). This was accomplished by spiking selected compounds into a heated air stream (plume generator). The output of the plume generator was then verified by continuously extracting samples from the plume generator stack and measuring the concentrations of the constituent gases using EPA Method 320. Simultaneously, the PFTIR remotely measured the concentrations of these gases immediately above the plume generator stack. The PFTIR results were compared to the EPA reference method data to assess overall performance of the PFTIR method. The third element of Phase I study involved using the PFTIR to measure actual emissions from a “well controlled” test flare at the John Zink Flare Test Facility on August 27, 2003. The primary objective of this test was to assess the logistical difficulties in making the flare measurements under actual field conditions. This test was conducted with the flare firing commercial grade propane at 10,000 lb/hr. The field of view of the PFTIR was moved to several locations in the flare plume to provide data at various zones within the plume. The data obtained from this limited test were useful in:

• Determining the signal levels in each of the analysis regions employed by the PFTIR

and assessing the overall signal to noise of the system;

• Determining typical gas temperatures to be expected in actual flare plumes;

• Determining the complexities of reducing the data obtained from an actual plume signal; and

• Assessing the concentrations of the major species that were present in an actual well-controlled flare plume, and calculating the associated combustion efficiency.

1.3 Data Quality Objectives For this program the TCEQ established specifications for data quality. These specifications were to:

1-4

• Determine combustion efficiency with a known certainty of ± 1%;

• Determine the gas concentration for individual species with an accuracy of ± 50%; and

• Speciate 90% of all flare constituents.

A Quality Assurance Project Plan (QAPP) dated August 6, 2003 was developed, and is included in Appendix A of this report. The purpose of the QAPP was to ensure that the PFTIR method was evaluated systematically, and that the data would be of the highest quality. The QAPP specified the elements of the Phase I test program and the data quality objectives. It also detailed the sampling and analytical procedures, as well as the statistical approach that would be used to report the results. 1.4 Principles of Passive Remote Sensing Using PFTIR Using traditional “active” open path absorption techniques to measure emissions from flare plumes would require transmitting a collimated beam of infrared light through a plume and then positioning a detector on the opposite side of the flare plume to detect the amount of energy absorbed by those compounds of interest. In this case, the specific wavelengths absorbed are indicative of the presence of specific compounds being present and the amount of light that is absorbed is proportional to the concentration of these species. The use of this approach is further complicated by the fact that the plume may change its direction of travel (relative to the light source) because of prevailing winds, thus requiring periodic re-alignment of the “active” light source and detector. While the use of “active” open path monitoring techniques might be used to characterize the emissions from some ground flares, it is impractical for use on elevated flares. An alternate approach, and the one employed in this study, is to use PFTIR for characterizing flare plumes. Unlike traditional spectroscopic methods, which rely on detecting the amount of light that is absorbed to identify and quantify the specie(s) present, the PFTIR operates on the principal of analyzing the amount of thermal radiation emitted by hot gases. In this case, the technique is “passive” since no “active” infrared light source is used. Rather, the hot gases of the flare become the infrared source and the PFTIR spectrometer is used to measure the amount of energy radiated from the flare plume. The use of PFTIR is possible because the IR radiation emitted by hot gases has the same pattern of wavelengths or “fingerprints” as the corresponding infrared absorption spectra. Consequently observing a flare from a distance with an IR instrument coupled to a receiver

1-5

telescope, allows for the rapid identification and quantification of the species in the flare plume. In this case, the signature arising from the hot gases is proportional to the concentration of the gas and to its temperature. Therefore, to conduct PFTIR measurements, the temperature must be deduced from the data, in addition to the concentration. This type of measurement also requires that the PFTIR be calibrated in absolute units of radiance (watts/cm2/ster/cm-1) using a black body radiation source. The intricacies of using PFTIR to measure flare emissions are described in more detail in Section 4 of this report. 1.5 Evaluating Simulated Flare Emissions from Plume Generator The performance of the PFTIR was evaluated by testing its ability to measure known emissions from a plume generator assembled by John Zink. During the simulated flare tests, certified gas mixtures (i.e., carbon dioxide, carbon monoxide, ethylene, butane, and propylene) were injected into John Zink’s “Big Blue” air preheater. The output of the preheater was ducted to a simulated flare stack that extended to a height of approximately 60 feet above ground level. The preheater was operated such that a constant volume of heated air was used to dilute the spiked gases, thereby simulating different combustion efficiencies. A photograph of the plume generator is shown in Figure 1-1. For this study, a test matrix with five different test conditions was developed which simulated three different target combustion efficiencies and two different exhaust gas temperatures. The matrix of test conditions for the plume generator is shown in Table 1-1. During each test sequence the output of the plume generator was sampled, and then analyzed using EFTIR (following EPA Method 320). The EFTIR was used to monitor the output concentrations of the plume generator, just below its exhaust plane.

1-6

Figure 1-1. Photograph of Plume Generator used to Evaluate PFTIR Photograph Courtesy of John Zink Company

“Big Blue” Portable Air Preheater

Plume Generator Exhaust Stack

1-7

Table 1-1. Target Test Conditions for Plume Generator Experiments

Test # Test Sequence Description Target

Temperature (oK)

Target Combustion

Efficiency (%)

1a, 1b Low Efficiency / High Temp. 498 96.0

2a, 2b Mid Efficiency / High Temp. 498 97.1

3a, 3b High Efficiency / High Temp. 498 98.7

4a, 4b Mid Efficiency / Low Temp. 423 97.1

5a, 5b High Efficiency / Low Temp. 423 98.7

Note: Each test sequence was repeated once. Test 1b was terminated early due to the depletion of spiking gas mixtures.

Simultaneously, the PFTIR receiver telescope was trained on the exit of the plume

generator stack, and was used to measure the radiant signature of the plume generator. The output concentrations for each of the target gas species were determined with the PFTIR. These were compared against the EFTIR measurements for each test sequence, the EFTIR values being taken as the reference, or “true” concentrations. A photograph of the PFTIR and associated telescope receiver is shown in Figure 1-2.

1.6 Summary of Key Observations and Conclusions Based on analysis of the Phase I Study results, a number of key observations and

conclusions can be made. A summary of these is provided below:

1.6.1 Analytical Study Findings The analytical study was completed shortly before deployment to the field in August 2003. Unfortunately due to compressed time schedules, the project team was unable to take full advantage of these findings before field deployment. The results do however; provide valuable information on peak radiance and the corresponding minimum detectable concentrations for those gas species that were evaluated as part of this study. The most significant findings were:

• The noise limited detection assessed for each compound implied a PFTIR measurement limit for calculated combustion efficiency of 99.95%;

1-8

Figure 1-2. Photograph of Passive Fourier Transform Infrared (PFTIR) Spectrometer

Photograph Courtesy of the John Zink Company

• The analytical study indicated that the Mercury Cadmium Telluride detector used during the Phase I Study was not sensitive enough at short wavelengths (~ 3µm). This would limit the detection of specific compounds such as THC compounds (>C5+), butane, and possibly propane;

• These results suggested that the use of an Indium Antimonnide (InSb) detector in tandem with the Mercury Telluride Cadmium (HgCdTe) (a sandwich detector) would substantially improve the detector performance at the short wavelengths (3 µm) by a

PFTIR Telescope Receiver

FTIR Spectrometer Housing

Plume Imaging Camera

1-9

factor of possibly one hundred. This would vastly improved the ability of the PFTIR to detect propane, butane, and THC compounds >C5+ in the C-H stretch region of 3000 cm-1; Unfortunately, obtaining such a detector was impossible between the time of the analytical study and the field test.

The analytical study provided valuable simulation data, which showed the sensitivity of

the radiant signal to sky background, path radiance, and the absolute calibration of the instrument. 1.6.2 Phase I Plume Generator Validation Study Findings The Phase I Plume Generator validation experiment was conducted August 25-27, 2003 at the John Zink test facility in Tulsa OK. During this study simulated flare emissions from the plume generator were measured with the PFTIR and the results were compared to measurements that were made simultaneously using an EFTIR. This series of experiments was used to assess the overall precision and accuracy (over a range of calculated combustion efficiencies and temperatures) of the PFTIR method, and to determine if the TCEQ data quality objectives could be met. The most significant findings are listed below.

• Overall, the results of the plume generator experiment suggest that the PFTIR is a potentially viable technology for determining selected species concentrations and combustion efficiencies in flare systems;

• The plume generator experiments provide an important set of ground-truth data for testing the PFTIR processing algorithms and their sensitivities to specific experimental measurements;

• Use of the EPA Method 320 EFTIR in concert with the plume generator was essential in generating “known” test conditions that could be used to accurately quantify the overall PFTIR measurement uncertainty, as well as to provide results which suggested refinements to the PFTIR instrumentation, operating procedures, and data processing algorithms;

• The output of the plume generator was independently verified by collecting a series of canister samples during Test 3a. The samples were then analyzed for each of the spiked compounds (i.e., carbon dioxide, carbon monoxide, ethylene, propylene, and butane) present in the plume generator stream. The results confirmed the accuracy of the Method 320 results.

• The overall analytical uncertainty of the calculated combustion efficiency for all test sequences using the PFTIR ranged from a low of 0.11% to a high of 0.82% (using Line-By-Line processing algorithm) and was within the project goal of +/- 1%.

1-10

• The calculated combustion efficiencies determined with the PTFIR were consistently low, and the biases ranged from -0.17 to -1.7 %. This can be partially explained by the low concentrations observed by the PFTIR for carbon dioxide using the LBL algorithm for some test sequences;

• The results of the precision test, Test 3b, showed that the PFTIR precision run-to-run was good overall; the coefficient of variation (CV) for the calculated combustion efficiency was 0.01% and the CVs for individual species ranged from a high of 29% for butane (known to be a problem because of spectral ranges used) to 5.4% for ethylene;

• The observed relative accuracies (RAs) for individual compounds (as measured by the PFTIR and EFTIR) generally were within +/- 50% for carbon dioxide, carbon monoxide, and ethylene. The RAs for butane and propylene were significantly higher than +/- 50% as was expected based on the lack of system response in C-H stretch region of 3000 cm-1;

• The absolute accuracy of the PFTIR in measuring individual species concentrations was limited by: 1) specific details of the quantification methods used to analyze spectra, 2) the absolute radiance calibration of instrument, and 3) the variable sky background and atmospheric path radiances which will need to be measured more frequently in future field studies;

• The results of the plume generator experiments are thought to represent poorer performance of the PFTIR than in an actual flare application for two reasons: 1) the temperatures observed in the actual flare plume were considerably higher than originally expected, and 2) temperature gradients were more significant in the plume generator than in the actual flare plume. These findings suggests that the PFTIR would have greater sensitivity in actual flare plumes; and

• Some unexpected disagreements were observed between PFTIR and EFTIR concentrations for individual compounds in specific tests. These are not completely understood, although the data does indicate that they are arising from the spectral reduction algorithms due either to the algorithm itself or to insufficient frequency of sky measurements and black body calibrations. Consequently, more effort is needed to refine and validate the data processing algorithms used to process the results from the PFTIR before attempting further field tests.

Comparisons of PFTIR and EFTIR concentrations indicated that significant bias was seen

for certain compounds in specific tests. The most striking of these were the low CO2 values that were seen in Tests 1a, 2a, and 3a. These tests were the higher temperature tests, and Test 1a was the lowest efficiency test. It was expected that these would be the easiest tests to analyze. To determine if the disagreements observed were inherent in the radiance data itself or a result of the analysis procedures used, an alternate method was employed to analyze the same set of radiance spectra. The method used was the Classical Least Squares (CLS) approach that was to be used

1-11

as an approximate real-time method to guide data taking in real time. CLS is more limited than LBL in the situations it can handle. Also CLS being a very rapid algorithm (a few tenths of a second as opposed to the 10 minutes or more required for LBL), allowed it to be optimized by altering spectral-analysis regions and/or spectral references to minimize spectral-residual errors and errors indicated by the EFTIR data. The known results were therefore used in optimizing the CLS method. Therefore, these results cannot be used to suggest that the CLS analysis provides for a more accurate measurement than LBL (it is expected that the CLS and LBL results should converge once all analytical issues are resolved). However, the fact that a single CLS algorithm could be set up to produce good agreement with the EFTIR data indicates that the issue is not the quality of the data collected. Instead, it suggests that the procedures used to reduce the data require further refinement. This should be the first step of any follow-on program.

1.6.3 Controlled Flare Experiment Findings

The PFTIR testing on the actual flare was performed on August 27, 2003. A series of measurements were made in which the field of view of the PFTIR was moved to a number of different locations in the flare plume. Overall, the results suggest that PFTIR should be capable of measuring the emissions from industrial flares under the rigors of actual field conditions. The following key findings were noted from the controlled flare experiment.

• The limited test of a controlled flare burning propane gas indicated that a high

combustion efficiency can be obtained. The combustion efficiencies calculated during the test ranged between 99.5% and 99.9%, thus suggesting that the flare was being operated for maximum combustion efficiency;

• The measured values for temperature, carbon dioxide, carbon monoxide, and propane from the PFTIR decreased with increasing distance above the flame as one would expect;

• The measured temperatures of the actual flare plume were higher at all positions above the combustion zone than previously thought. This indicates that PFTIR measurements should not be sensitivity or signal to noise limited;

• No structured plume gradients were observed anywhere above the flame. Instead an inhomogeneous distribution of gas concentrations and temperatures was observed with both the PFTIR and thermal imager. This should be an easier scenario to treat when measuring emissions from properly controlled industrial flares;

• The flare test provided data on the true signal intensities for a high efficiency flare plume; and

1-12

• The controlled flare experiment provides valuable information for assessing the logistical difficulties that can be expected when measuring flare emissions with the PFTIR under actual field conditions.

1.7 Path Forward Recommendations The results of the Phase I testing suggest that PFTIR can be a viable approach for continuous near-real-time measurement of flare emissions and calculation of flare efficiencies. This implies that a second phase is warranted to continue the evaluation and development of PFTIR for field measurement of flare emissions. Based on the results of the Phase I Study it appears that further effort is first needed to refine the analytical methodology used to process the PFTIR data. Once the data processing algorithms are refined and tested, to the extent possible with Phase I data, an abbreviated plume generator experiment would be desirable to verify the effectiveness of the refinements made to the data processing algorithms, as well as to assess the improvements provided by the PFTIR detector upgrades. Following this, a second series of tests should be conducted on flares in which the gas feed streams can be well instrumented (possibly the ground-level flare at John Zink). Such a test would allow for the incorporation of the lessons learned from Phase I, and provide further documentation of the ability of PFTIR to accurately quantify the emissions from the complex gas composition of real flares in realistic environments. These tests should then lead to field measurements on flares at industrial facilities in Texas.

2-1

2.0 Introduction This report documents the Phase I field testing program conducted for the UH and TCEQ

to evaluate a method for determining the combustion/destruction efficiency of operating process flares. The results from this study will be used to guide and develop future tests of actual flares. This section provides background information and the objectives of the study, along with the organization of the report. 2.1 Background

The Houston/Galveston area (HGA) has been designated as “non-attainment” for achieving the National Ambient Air Quality Standard (NAAQS) for ozone. The mechanism for the formation of ozone (and subsequently, photochemical smog) is based on reactions of volatile organic compounds (VOCs) and oxides of nitrogen (NOx) in the presence of sunlight. TCEQ has combined these reactions with emissions and meteorological data to develop a sophisticated model for predicting the ambient ozone concentration across the HGA.

In an effort to bring the HGA into compliance with the NAAQS, TCEQ has promulgated

rules under the State Implementation Plan (SIP) for monitoring and control of both NOx and VOCs. These rules have been based on modeling results, which suggested that the controls prescribed in the rules would be sufficient to attain the NAAQS. However, recent data have suggested that the actual emissions of VOCs have historically been grossly under-estimated and reported. Furthermore, there appear to be certain under-reported species, such as olefins, that may be considered as “highly reactive” in the formation of ozone.

Based on input from a group of experts, one likely source for these under-reported highly

reactive VOC emissions is the great number of process flares that are operated in the HGA. Emissions from process flares are very difficult to monitor, compared to other combustion sources. This is because the flare tip is elevated, and the flame typically extends into the open air where the emissions are rapidly dispersed. This configuration negates the use of conventional sampling methods. As a consequence, there is a limited amount of data available that describes the emissions from actual flares. These data have been used to establish criteria for minimum heating values of the gases being flared, and maximum flare tip velocity, to ensure that the destruction efficiency of the flare achieves acceptable levels. However, sufficient data exist to suggest possible, atypical operating modes (such as over-steaming), in which the flare’s destruction efficiency could be reduced well below accepted values.

2-2

As part of its ongoing efforts to attain the ozone NAAQS, TCEQ is currently pursuing a number of initiatives to improve the quality of the data that are being used for modeling ambient ozone in the HGA. In the present study, TCEQ and UH are working together to characterize the combustion/destruction efficiency of operating flare systems and to develop methods for the more accurate determination of mass emissions from operating flares.

The University of Houston contracted URS Corporation (URS) to evaluate the feasibility

of Passive Fourier Transform Infrared (PFTIR) spectroscopy as a method for estimating flare destruction efficiency of volatile organic compounds (VOCs), and the total speciated mass emissions from a flare. URS assembled a team of scientists and engineers with extensive experience in both conducting and evaluating measurements of this type. The team includes Industrial Monitoring Applications Corporation (IMACC) and John Zink Company (John Zink).

In cooperation with TCEQ and UH, the URS team developed a multi-phase study for the

purpose of determining whether PFTIR is a viable method for measuring mass emissions of individual species and total hydrocarbon emissions from process flares. The first phase of the study was to evaluate the ability of the PFTIR to measure concentrations in a controlled test environment. The second phase would be to evaluate the PFTIR technology on actual process flares operating in the HGA. This report documents the results of the Phase I testing.

A Quality Assurance Project Plan (QAPP) was developed to ensure that the PFTIR

method was evaluated systematically, and that the data would be of the highest quality. The QAPP specified the elements of the Phase I test program and the data quality objectives. It also detailed the sampling and analytical procedures, and the statistical approach that would be used to report the results.

The Phase I test program was comprised of three major elements. The first of these was

an analytical study of the PFTIR to determine the expected signal strength, minimum detection limits, and the sensitivity of the signal to plume gradients. The second element was a field test in which a simulated flare plume was generated containing known concentrations of selected gases. These gases were emitted from a stack from which a sample was extracted and measured using an EPA-approved reference method. The PFTIR then measured the concentrations of these gases immediately above the stack, and the results were compared with the reference method data. In the third element of Phase I, the PFTIR was used to measure emissions from an actual flare at a test facility, to develop information on the logistics of making the measurement.

2-3

2.2 Objectives As described above, TCEQ’s ultimate goal is to improve the quality of the emissions data

that are being used as part of the ongoing effort to achieve compliance with the ozone NAAQS. As part of that effort, TCEQ would like to develop a method to quantify the mass emissions of VOCs from process flares. In order to meet those goals, the overall objectives of Phase I were to:

1. Evaluate the ability of the PFTIR spectroscopic methods to determine combustion/destruction efficiency and total speciated mass emissions from operating process flares within a known level of accuracy and precision; and

2. Determine combustion/destruction efficiency and total speciated mass emissions in

combination with speciated waste gas measurements for flare systems under a variety of operating conditions.

To meet these objectives, a combined test plan/quality assurance project plan (QAPP) was developed to evaluate the ability of the PFTIR to both identify and quantify speciated emissions from a flare under simulated field conditions. The QAPP described the three major tasks that would comprise the Phase I testing program. These consisted of the following:

• An analytical evaluation was conducted to calculate the minimum detectable gas concentrations for the gas species used in the test program, and from these, the maximum combustion efficiency possible using the PFTIR;

• A series of tests were performed on a simulated flare plume to measure simulated flare emissions while operating under actual field conditions; and

• A test of an actual flare at the John Zink flare test facility was conducted to calculate combustion/destruction efficiency under actual field conditions.

2.3 Organization of Report

The remainder of this report provides a description of the test facility (Section 3), description of the sampling and analytical procedures that were used (Section 4), a discussion of the results (Section 5), and conclusions and recommendations (Section 6). The Quality Assurance Project Plan (QAPP) is presented in Appendix A, and Appendices B through F provide the data that were collected and analyses that were performed.

3-1

3.0 Test Facility Description and Process Operation

Field testing for Phase I of the test program was conducted at the John Zink test facility located in Tulsa, OK. The first series of tests were conducted using the “big blue” air heater system, which provided a heated ambient air stream that was subsequently spiked with volatile organic compounds (VOCs) to produce a simulated flare plume. The second series of tests were conducted using an industrial size test flare fired with commercial-grade propane. Details of these systems are described below, followed by a discussion of process operation and meteorological conditions during the test periods.

3.1 Plume Generator System The second task in the Phase I testing employed a plume generator system, which

contained a portable air heater that delivered approximately 275,000 standard cubic feet per hour (scfh) of preheated air through a duct. This preheated air was subsequently spiked with the constituent test gases: carbon dioxide, carbon monoxide, butane, ethylene, propylene, and propane. The heated air containing constituent gases was routed to a 36-inch, refractory-lined stack and discharged at a height of 65 ft to provide a simulated plume suitable for passive Fourier transform infrared (PFTIR) analysis. Figures 3-1 and 3-2 provide schematic diagrams of the plume generator test apparatus.

Test equipment for the plume generator tests consisted of the air heater, stack, spike gas storage and feed systems, and associated controls and instrumentation. Spiking occurred after the air heater through a port on the air heater outlet duct upstream of the plume stack. Spiking gas inventories were estimated prior to the test to provide enough of each spiked gas for approximately 5 to 10 minutes of sampling for each of the planned tests, based on the maximum specified concentration for each constituent. Carbon dioxide, carbon monoxide, butane, ethylene, and propylene were metered into the heated air from high-pressure certified gas cylinders containing 3-5 volume percent mixtures in nitrogen. In addition, as a result of a field modification to the plume generator test plan (discussed in Section 5 of this report), propane was substituted for butane in one of the test conditions because supplies of butane had been exhausted in earlier tests. Commercial grade propane for this test was supplied from the burner test area fuel supply network, and was routed to the mixing manifold via existing in-facility piping. The source of the propane was the propane liquid bulk storage tank⎯the same tank used to supply the propane for the controlled flare test.

3-2

Control instrumentation for the plume generator system consisted of the following, as shown in Figures 3-1 and 3-2:

• Pressure transmitters for process streams;

• Differential pressure transmitters for orifice flow meter measurements;

• Needle valves for fine flow adjustment of spiking gas flows;

• Temperature transmitters for thermocouples on gas stream temperatures;

• Orifice flow meters for spiking gases; and

• Nozzle flow meters for plume generator air flow measurement.

The air heater utilized a gas-to-gas shell and tube heat exchanger. The shell fluid was hot gas, which generated by direct firing of natural gas in a burner that was located inside the air heater. The ambient air (cold gas) passed through on the tube side, and was heated to the desired temperature by the hot gas. The flow rate for the heated air was monitored and controlled at the ambient air inlet to the “big blue” air heater unit, and this flow rate was used to calculate the spiking rates necessary at the air heater outlet location to achieve the target concentration in the simulated plume gas exiting the stack. An extractive FTIR system provided real-time analysis of moisture, carbon dioxide, carbon monoxide, butane, ethylene, propylene, and propane in a continuous slip stream of the simulated plume gases collected from a sample port located at the 62-ft level of the stack.

Prior to the start of testing, a smoke test was performed in which a smoking device was placed at the ambient air inlet of the air heater. The purpose of this test was to determine if there were any leaks in the ducting system or stack that were being used for the plume generator tests. At that time, smoke was observed exiting from the air heater hot gas exhaust. This suggested that some of the cold ambient air was leaking into the hot gas within the heat exchanger, and passing out the hot gas exhaust. Since this leak occurred downstream of the location that the ambient air flow was being measured, the actual flow of heated air into the plume generator was less than what was being measured. Therefore, the predetermined flow rates of spiked gases that were injected downstream of the air heater produced higher concentrations of spiked gases in the plume than were targeted in the QAPP. As a result, the spiked gas concentrations calculated from the flow meters at the site were not used as another method to determine the accuracy of the PFTIR concentration measurements. The extractive FTIR measurements that were made according to EPA Method 320 were used as the “true” emissions values to determine accuracy of the PFTIR.

3-3

3.2 Controlled Flare System The third task in the Phase I testing was conducted on a flare system that is part of John

Zink’s ongoing flare development program. Unlike the plume generator tests, no VOC gas spiking was conducted during the controlled flare test, and no EFTIR sampling was performed. Figure 3-3 is a schematic diagram of the controlled flare system used for the tests. John Zink personnel operated this equipment and accompanying instrumentation during the test. Control instrumentation consists of the following:

• Pressure transmitters for process streams;

• Differential pressure transmitters for orifice flow meter measurements;

• Flow control valves for fuel; and

• Temperature transmitters for thermocouples on gas stream temperatures.

Propane for the test was fed from Zink’s on-site liquid storage to the evaporator coil in a

hot water bath liquid petroleum gas (LPG) vaporizer. The water in the vaporizer was heated by firing a natural draft, natural gas burner into a fire tube submerged in the water bath. The propane liquid was vaporized in a separate pipe coil that was also submerged in the water bath. Combustion products from the gas burner heater flowed through the fire tube and exited to the atmosphere via a stack.

The vaporized propane flowed out of the vaporizer to a gas flow measurement and flow

control station. A pressure vessel was connected to the line carrying the propane gas to increase the system volume as an aid to pressure regulation and flow control. A pressure control valve (PCV-3110) provided a constant pressure to the inlet side of flow measurement orifice (RO-3112). A pressure transmitter (PT-3110) and temperature sensor/transmitter (TT-3111) provided information regarding gas conditions upstream of the orifice. A pressure transmitter (DPT-3112, 2.6 inch diameter) provided the differential gas pressure across the orifice and was used to maintain the gas flow rate using the flow control valve (FCV-3112). The propane gas then flowed to the air assisted flare where it was ignited by pilots and burned. For the controlled flare test, a target propane fuel flow rate of 10,000 lb/hr was established for the PFTIR testing period.

Smoke formation in the flare was suppressed by forced mixing with air. The air for smoke suppression was supplied by a blower, and air velocity was measured by a Pitot tube in the vertical stack section between the blower and the flare burner.

3-4

Figure 3-1. Schematic Diagram for the Plume Generator System

3-5

Figure 3-2. Schematic Diagram for the Plume Generator System (Propane, CO and CO2 injection test)

3-6

Figure 3-3. Schematic Diagram for the Controlled Flare System

3-7

3.3 Process Operation During Testing The plume generator and controlled flare were operated by John Zink personnel during

the testing. During the testing, the operating data for these processes were collected and stored electronically on a data historian. These data are discussed in the following sections.

3.3.1 Plume Generator A single set of values for the plume generator process parameters was recorded

electronically by John Zink personnel during each test condition once steady-state plume VOC concentrations were achieved (as measured by the EFTIR system). Plume generator system process settings for each test condition are summarized in Table 3-1. Detailed process data and additional information on spiking gas cylinder compositions are provided in Appendix D.

Total air flow rates downstream of the gas spiking location ranged from 272,000 to

279,000 scfh for the various tests, resulting in an exit velocity at the stack of approximately 16 to 19 ft/sec. Ambient temperatures ranged from 86°F to 97°F on 8/25/03 and 8/26/03, with slightly lower ambient temperature (76°F) for the tests on 8/27/03. Winds were generally from the South/Southeast at speeds ranging from of 5 to 11 miles per hour (mph) over the three-day test period.

Loss of heated air was noted across the air heater system during the initial smoke testing

of the air heater system due to a leak in the heat exchanger. This leak resulted in heated air flow rates at the outlet of the air preheater that were approximately five percent lower than the measured inlet air flow rates to the air preheater. Since spiking rates for target compounds were established based on the measured inlet flow rate to the air heater, the leakage resulted in species concentrations that were somewhat higher than the target concentrations. This was further magnified in the case of the target CO2 concentration, which was also affected by the significant levels of CO2 (~400 ppmv) that were present in the ambient air. As a result, the flow data for the gas flows that were used to calculate the spiked gas concentrations were not included in the statistical analysis of the final results.

3-8

Table 3-1. Plume Generator and Spiking System Process Settings

Test Condition

Date

Time Am

bien

t Tem

p (F

)

Rel

ativ

e H

umid

ity

(%)

Win

d Sp

eed

(mph

)

CO

2 Flo

w (s

cfh)

CO

/N2 F

low

(scf

h)

But

ane/

N2 F

low

(s

cfh)

Eth

ylen

e/N

2 Flo

w

(scf

h)

Prop

ylen

e/N

2 Flo

w

(scf

h)

Prop

ane

Flow

(scf

h)

Exh

aust

Gas

T

empe

ratu

re (°

)

Inle

t Air

Flo

w

Noz

zle

Del

ta P

(in.

W

C)

Inle

t Air

Flo

w a

(scf

h)

Tot

al G

as F

low

b (s

cfh)

Stratification 8/25/03 3:59:44 97 30 8 0 70 0 0 0 0 438 2.45 272,226 272,296Startup 8/26/03 9:45:18 85 53 6 1671 0 0 0 0 0 439 2.42 273,518 275,189Test 1a 8/26/03 9:56:39 86 52 7 1667 785 758 0 0 0 434 2.44 274,394 277,604Test 2a 8/26/03 10:25:50 88 46 5 1660 396 377 150 256 0 435 2.43 273,331 276,170Test 3a 8/26/03 10:49:25 89 44 6 1662 119 152 91 153 0 437 2.46 274,762 276,938Temp. change 8/26/03 12:47:22 93 35 9 1661 0 0 0 0 0 311 2.49 275,431 277,092Test 4a 8/26/03 1:09:23 94 34 5 1659 396 381 151 257 0 309 2.45 272,962 275,806Test 5a 8/26/03 1:39:13 94 35 8 1670 121 150 90 151 0 307 2.48 274,629 276,810Test 4b 8/26/03 2:15:11 96 33 11 1653 391 0c 164 255 0 307 2.47 273,581 276,043Test 5b 8/26/03 2:54:04 95 33 9 1665 122 151 93 150 0 307 2.48 274,381 276,562Test 2b 8/26/03 3:45:13 97 28 6 1666 395 382 151 255 0 426 2.42 270,554 273,404Test 3b 8/26/03 4:09:37 96 32 7 1666 124 150 94 152 0 440 2.44 271,914 274,099Test 1b 8/27/03 10:46:52 76 72 6 1657 555 0 0 0 24 450 2.44 276,944 279,179Heated air only

8/27/03 11:33:17 76 72 6 0 0 0 0 0 0 443 2.46 278,076 278,076

a Measured at the inlet to the air preheater system. b Total gas flow calculated based on the measured inlet rate plus the spiking gas rates. Does not account for losses of heated air across the preheater due to leakage. c Butane supply from the first set of cylinders had been exhausted by this time during the test. Additional butane cylinders were connected to the spiking system for the subsequent tests.

3-9

3.3.2 Controlled Flare Figure 3-4 provides a graphical summary of key process parameters and meteorological

conditions during the controlled flare test. A statistical summary is provided in Table 3-2 based on the process data collected by the John Zink data acquisition system (DAS) at 1-second intervals. Prior to the controlled flare test, John Zink conducted a series of short duration flare burns (<20 seconds) to observe flame characteristics and wind influences to aid in the selection of an appropriate fuel feed rate for the PFTIR test. Based on these observations, the time necessary to obtain PFTIR data, and the amount of fuel available for the test run, a nominal propane fuel flow rate of 10,000 lb/hr was selected by the project team. This rate was selected to provide a flame and plume profile that were considered representative “real world” flare conditions. The average propane fuel flow rate to the flare during the PFTIR test was 9,960 lb/hr with a standard deviation of 80 lb/hr. The air-assist blower settings were held constant during the test to supply air at a rate of 603 standard cubic feet per second (scf/sec) as measured by a pitot tube in the air supply duct to the flare. Resulting flare tip velocities were in the range of 40 to 50 feet per second.

Table 3-2. Statistical Summary of Process and Meteorological Data for

Controlled Flare Test (5:12 PM to 6:03 PM) Parameter Average Minimum Maximum Std. Deviation

Fuel flow rate (lb/hr) 9,960 9,775 10,655 80 Fuel flow orifice differential pressure (inches water)

296 289 344 6.8

Fuel temperature at orifice inlet (°F)

45 33 108 15

Fuel pressure at orifice inlet (psig)

10.9 10.6 13.8 0.3

Ambient temperature (°F) 93.5 92.6 95 0.7 Ambient relative humidity (%)

37.5 34.3 39.9 1.5

Ambient pressure (psia) 13.25 13.25 13.25 0 Wind speed (mph) 5.7 1.8 11.8 1.5

Figure 3-4 also shows the meteorological conditions during the test period. Ambient air temperature and relative humidity remained relatively constant during the test with average values of 93.5ºF and 37.5 percent relative humidity (RH), respectively. Skies were generally partly cloudy during the test period with an average wind speed of 5.7 mph. Winds generally blew from the southeast to the northwest (~330 degrees), with occasional periods where the wind was blowing to the north/northeast (approximately 0-30 degrees) across the flare test pad. Detailed process data are provided in Appendix D

3-10

Figure 3-4. Controlled Flare Process and Meteorological Data During the 8/27/03 Test Period.

9200

9400

9600

9800

10000

10200

10400

10600

10800

17:12:0017:14:0917:16:1817:18:2717:20:3617:22:4517:24:5417:27:0317:29:1217:31:2117:33:3017:35:3917:37:4817:39:5717:42:0617:44:1517:46:2417:48:3317:50:4217:52:5117:55:0017:57:0917:59:1818:01:27

Time on 8/27/03

Fuel

Flo

w R

ate

0

10

20

30

40

50

60

70

80

90

100

Met

erol

ogic

al. D

ata

Para

met

ers

Fuel Flowrate, lb/hrAmbient Temperature, degFAmbient Pressure, psiaWind Speed, mphAmbient Humidity, %RH

Ambient Temperature, F

Propane Fuel Rate, lb/hr

Ambient Humidity, %RH

Ambient Pressure, psia

Wind Speed, mph

4-1

4.0 Sampling and Analytical Procedures This section describes the sampling and analytical procedures used during the plume generator and controlled flare tests. The purpose of the plume generator study was to determine accuracy and precision of the total PFTIR inversion process. Measurement of actual gas concentrations at the output of the plume generator allowed for direct comparison of the PFTIR data with EFTIR data as well as with selected canister grab samples. EFTIR and canister results represent what is believed to be the actual concentrations. This allows for a direct computation of system accuracy. Repeated measurements being performed on the most stressful case during the plume generator test series (test condition 3b) provided a measurement of precision of the process. The purpose of the controlled flare tests was to determine the logistical challenges of collecting PFTIR data from an actual flare under “normal” ambient conditions. 4.1 Sampling Procedures

The following section describes the sampling procedures for each of the measurement methods that were followed as part of the test program. Along with the PFTIR measurements, three other sampling methods were identified for use as reference methods, so that the “true” value of the emissions could be calculated with a known certainty. This “true” value then compared to the PFTIR results. These methods include the flow rates for the heated air and spiked gases, EFTIR, canisters, and PFTIR. As mentioned above, the results of the flow meter, EFTIR, and canister measurements were used to establish the “true” value for the spiked gas concentrations, so that they could be compared to the PFTIR. The canister and PFTIR sampling procedures for the plume generator and controlled flare tests are discussed separately. 4.1.1 Airflow Metering and Gas Spiking – Plume Generator Tests

The airflow to the plume generator system was measured using a calibrated flow meter located upstream of the variable speed blower at the inlet to the air heater system (refer to Figure 3-1). The air heater inlet air flow rates during the plume generator tests were controlled between 272,000 and 278,000 standard cubic feet per hour (scfh) by adjusting settings on the blower. Constituent gas spiking rates and gas manifold flow control settings required to achieve the target constituent gas concentrations during each test were determined prior to the testing. Constituent gases were measured by calibrated flow orifices, as described previously in Section 3. At the beginning of each day of testing, the plume generator air-heater system was started and allowed to come to steady state at the testing stack temperature. The target constituent

4-2

concentrations and stack gas temperatures for each test are summarized in Table 4-1. Plume generator system instrument calibration data are provided in Appendix E.

4.1.2 Extractive FTIR (EPA Method 320) - Plume Generator Tests

EFTIR samples were obtained to determine the concentration of the spiked gases in the exhaust from the plume generator. Figure 4-1 is a schematic of the EFTIR sampling system. A fully integrated extractive MKS FTIR system was used. The sampling system comprised a stainless steel probe, a 100-ft heated (160°C) 3/8-inch O.D. PFA-grade Teflon extraction line, an MKS FTIR spectrometer interfaced to a heated (150°C) nickel-coated sample cell, a sample pump, and a flow regulating rotameter. A diaphragm pump at the outlet of the EFTIR instrument was used to continuously draw a metered quantity of the simulated stack gas mixture through a sample line from the center of the plume generator stack, using a sample port located at approximately 62 ft of the 65-ft stack. The sample stream passed through the sampling cell for analysis according to EPA Method 320. Sample flow was maintained at approximately 7 standard liters per minute (slm).

Cell pressure vacuum was continuously recorded during measurement periods using a pressure sensor calibrated over the 0 – 900 torr range. The pressure is required in the quantification of the spectral data. An in-line filter, prescribed by EPA Method 320, was not used in the sample extraction line since the particulate loading was insignificant. The absence of the in-line filter did not affect sample results.

The EFTIR system operated continuously during each day of testing real-time analysis of the plume stack composition based on sample measurements generated at 42-second intervals. A gas spiking “tee” in the extraction line, as depicted in Figure 4-1, was used to inject certified gas standards as specified by EPA Method 320 validation procedures.

4.1.3 Canisters

For verification of the concentration of the gas species of interest, grab samples were collected during the plume generator and controlled flare tests. In the plume generator tests, the grab sample results were used as a cross-check of the EFTIR data. For the controlled flare test, grab samples of the gas being fed to the flare were analyzed to verify the composition provided by the gas supplier. The procedures used to collect the canister samples for both the plume generator and controlled flare tests are described in the following paragraphs.

4-3

Table 4-1. Target Test Conditions for the Plume Generator Tests Stack Gas Concentration (ppmv)

Test No. Description

Stack Temp

(K) CO CO2 H2O Butane Ethylene Propylene

Target Simulated Combustion

Efficiency (%)

1a, 1b Low Efficiency / High Temp. 498 130 5500 Ambient 100 0 0 96.0

2a, 2b Mid Efficiency / High Temp. 498 65 5500 Ambient 50 25 25 97.1

3a, 3b High Efficiency / High Temp. 498 20 5500 Ambient 20 15 15 98.7

4a, 4b Mid Efficiency / Low Temp. 423 65 5500 Ambient 50 25 25 97.1

5a, 5b High Efficiency / Low Temp. 423 20 5500 Ambient 20 15 15 98.7

4-4

Figure 4-1. Schematic of Stack Gas Extractive FTIR System

EFTIR Sample Cell 150°C

Vacuum Pump

Valve

Rotameter

Exhaust

Valve

Calibration Spiking T

Certified Standard Calibration Gas(es) Stack

Heated Extraction Sample Line

Sampling Probe Valve

Canister sample port

4-5

4.1.3.1 Plume Generator Tests Grab samples of the plume generator stack gas were collected using 1-liter Summa

stainless steel canisters. Canisters were supplied by Air Toxics, LTD (ATL) under an initial vacuum of approximately 28 inches mercury (in. Hg). Samples were collected from a sampling post at the inlet of the EFTIR sample cell as shown in Figure 4-1. The initial vacuum of each sample canister was verified prior to sample collection using a vacuum gauge supplied by ATL. The sampling valve was then purged with sample gas, the valve was closed, and the canister was connected to the port. The sampling valve was then opened, and the needle valve on the canister was used to control the flow of sample gas into the canister until a final vacuum reading of approximately 10 in. Hg was obtained. A slight negative pressure was maintained in the canisters to prevent moisture condensation. Table 4-2 summarizes the canister samples collected during the plume generator tests. Three canister samples were collected during test condition 3a on 8/26/03, and a fourth sample of UHP grade nitrogen was collected on 8/26/03 as a field blank to assess background levels relative to sample analytical results. Three replicate samples were collected to provide a sufficient number of samples to allow for statistical analysis. Test 3a was chosen because it included all of the species of interest at their lowest concentrations, which is considered to be a more difficult test for the PFTIR.

Table 4-2. Canister Samples Collected During the Plume Generator Tests

Canister ID Test Condition Date, Time Initial/Final Pressure

(in Hg) Test 3-A Test 3a 8/26/03, 10:49 -28 / -10 Test 3-B Test 3a 8/26/03, 10:52 -28 / -10 Test 3-C Test 3a 8/26/03, 10:55 -28 / -9 Test 3-D Field Blank a 8/26/03, 10:59 -28 / 0

a A sample of UHP grade nitrogen was collected as a field blank. 4.1.3.2 Controlled Flare Tests

Grab samples of the propane fuel fed to the flare were collected using 1-liter Summa stainless steel canisters during the controlled flare test burn. Canisters were supplied by ATL under an initial vacuum of approximately 28 in. Hg. Samples were collected from a dedicated sample port/valve located on the underside of the fuel supply line to the flare. The initial vacuum of each sample canister was verified prior to sample collection using a vacuum gauge. The valve was then purged with propane gas, the valve was closed, and the canister was connected to the sample port. The sample port valve was then opened, and the needle valve on

4-6

the canister was used to control the flow of sample gas into the canister until a final vacuum reading of approximately 10 in. Hg was obtained. A slight negative pressure was maintained in the canisters to prevent moisture condensation. Table 4-3 provides a summary of the canister samples collected during the controlled flare test. The first canister sample was collected on 8/27/03 at approximately 15 minutes after the start of the controlled flare test, and a second sample was collected 20 minutes later.

Table 4-3. Canister Samples Collected During the Plume Generator Tests

Canister ID Date, Time Initial/Final Pressure

(in Hg) Test 6-1 8/27/03, 17:30 -28 / -10

Test 6-2 8/27/03, 17:50 -28.5 / -10

4.1.4 Passive FTIR PFTIR sampling procedures for the plume generator and controlled flare tests are

described below, and include descriptions of the location of the PFTIR relative to the source, the fields of view within the plume that were collected, and any deviations from the test plan that occurred during the testing. A photograph of the PFTIR telescope is shown in Figure 4-2.

4.1.4.1 Plume Generator Tests Device Location Relative to Source – The passive FTIR telescope was placed approximately 115 feet (35 m) from the exhaust stack in a position that would allow for erecting and securing a shelter to be used to shield the instrument computer from the sun and heat. The position chosen provided a clear view of the plume without interference from adjacent stacks, and it was at approximately 90 degrees from the prevailing wind direction. This position allowed for a clear view of the exhaust even if winds were high enough to cause a bending of the plume above the stack. Figure 4-3 provides a view of the plume generator stack from the spot where the PFTIR was located.

4-7

Figure 4-2. Photograph of Passive Fourier Transform Infrared (PFTIR) Spectrometer

Photograph Courtesy of the John Zink Company

Plume Imaging CameraFTIR Spectrometer

Housing

PFTIR Telescope Receiver

4-8

The distance from the stack was a compromise between range and angle of view. The

signal received by the FTIR will reduce as the square of the distance from the source. For this

first test series, a high signal to noise was desired; therefore, a shorter distance was preferred.

This increased the angle of view somewhat as shown in Figure 4-4. The primary influence of the

angle of view was to increase the amount of the plume that would be seen above the exit plane of

the stack. For the distance chosen, Figure 4-5 shows that the 12 in. diameter field of view of the

instrument began at the edge of the stack on the front and ended at 9.5 in. above the stack at the

rear. This put the maximum height of the field of view at 21.2 inches at the rear edge of the

stack. This was considered acceptable because it was believed that ambient air mixing would not

be severe at this height above the exit plane.

Fields of View – The data that are most useful for computation of concentrations are those made at the centerline of the plume with the plume filling the field of view (FOV) of the PFTIR. If the FOV is not filled, a true radiance measurement cannot be made. Data collected off of centerline were used to estimate horizontal or vertical gradients in the plume, which, if significant, were included in the data reduction calculations. As a result, the first pass through the test series (the “a” series tests) concentrated on centerline measurements. This assured that sufficient data were available, in the event of a fuel or time shortage, to allow for rigorous testing of the data reduction algorithms. Subsequent tests (the “b” series tests) were performed using coarse and fine traverses of the plume, in which the FOV was traversed across the width of the plume. This allowed for a comparison of the results (in Section 5) that were obtained using just the centerline data compared to the data obtained at different locations in the plume. The results of the tests in which the FOV was traversed across the width of the plume were also used to estimate concentration and temperature gradients. Figure 4-6 shows two fine traverses performed on 8/26/03 from 16:14 to 17:10. The first point at position zero was at the center of the stack. The next two points were roughly equally spaced from center to the right edge of the stack. The last two points were actually off the right edge of the stack by roughly 25 percent and 50 percent of the field-of-view. Position 2 was the position that could be most accurately

4-9

Figure 4-3. Photograph of Plume Generator used to Evaluate PFTIR Photograph Courtesy of John Zink Company

Plume Generator Exhaust Stack

“Big Blue” Portable Air Preheater

4-10

Figure 4-4. Angle of PFTIR Line of Sight at the Stack

Figure 4-5. Relative Position of PFTIR Field of View Across the Top of the Stack

4-11

placed since the right edge of the field-of-view just touched the right edge of the stack. The next most accurate was the center line position since it was located by moving half way between the two edges of the stack as measured on the PFTIR tripod vernier. The other points were more difficult to repeatedly locate because of limits in the resolution of the vernier. The scatter in the data is probably indicative of this repositioning accuracy, as well as wind effects. The first three points were all within the stack diameter and positioned, as much as possible, evenly from centerline to the right edge of the stack. These points show little if any gradient from center to edge. It was therefore concluded that the simulator stack essentially had a “top-hat” profile with a rapid drop off outside the stack diameter. Deviations from the Test Plan – The only significant alteration to the original test plan was the decision to not traverse the FOV across the width of the plume during the “a” series of tests. It was decided to concentrate on centerline measurements only (as opposed to

0

0.2

0.4

0.6

0.8

1

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5

Position (arb. Units)

Rel

ativ

e Si

gnal

Inte

nsity

Figure 4-6. Observed Signal Profiles Across the Simulator Stack

4-12

measurements in which the FOV was traversed across the width of the plume) during the first run through the test series (the “a” series tests) conducted on August 26. By not traversing during this first test series, many more spectra were collected on the centerline for each simulated condition. This provided a more comprehensive data set for testing the methods and their precision and allowed for longer term averaging of data at centerline. The second run through the test series, performed on the afternoon of August 26 and the morning of August 27, was conducted using FOV traverses (in the “b” series tests) to determine if any gradients needed to be treated in the model. One unforeseen problem that was encountered during testing was the overheating of the PFTIR control computer. The computer used in this test had a very high-speed processor (~3 GHz) which runs much hotter than slower processors. This computer was used so it could collect data rapidly and also process data in near-real time if this was required in the field. Because the ambient temperatures were high during testing, this computer started to fail when used without exterior cooling. Once a large fan was provided by John Zink and the case of the computer opened to allow for air circulation, this problem disappeared. Later in the testing, John Zink provided a small air conditioned trailer that was used to house the computer and its operator. 4.1.4.2 Controlled Flare Tests

Device Location Relative to Source – For the controlled flare tests, the PFTIR was placed at a location that would allow for observation of the plume at near right angles with either of the two dominant wind directions observed the day of testing. This position was somewhat limited in total path-length because of structures at the Zink facility. As a result, the position ultimately used was the longest distance that could be arranged with the PFTIR at the proper location for the dominant wind fields. Wind speed was high enough to deflect the exhaust plume axis from vertical as it left the flare stack. This meant that the PFTIR was looking through a more horizontal portion of the plume. The angle at which the FTIR viewed the plume (the view angle) was therefore less significant than it was for the simulator tests where the plume was vertical. In section 4.1.4.1, it was shown that the PFTIR view angle caused the FTIR to “see” gases higher above the stack exit plane at the rear of the stack. This will be a less significant effect if the PFTIR is measuring down wind of the stack and viewing a more horizontal plume.

4-13

Fields of View – Thermal imagery was used to assist in positioning the PFTIR FOV in the flare plume. The zone monitored needed to be above the combustion zone so all chemical reactions were complete, yet low enough so the temperature would be above 150 to 200°C to optimize detection. The FOV began at about 1 meter above the visible flame and was increased to higher positions during testing. Table 4-4 shows the sequence of events during testing, as well as the times chosen for center-line averaging at each height and the temperatures deduced from the line-by-line algorithm in each case. The position at 17:25 was just above the flame and some combustion occurred in the FOV of the PFTIR. The position set at 17:28 was higher but occasionally “flamelets” could be seen in the PFTIR FOV. The highest position, at 17:41, was well above the combustion zone and no IR-visible flames appeared in the PFTIR FOV. This was also the lowest temperature position as would be expected. Data from this position was used for most of the combustion efficiency estimates because it was considered the most representative of actual products emitted to the air. At 17:55, the FOV was again adjusted downward to the edge of the combustion zone itself and this position provided the highest observed temperature. Horizontal traverses of the FOV were done at the various heights, as indicated in the table. The plume position continuously shifted due to the wind, so an exact position relative to the centerline was difficult to define.

Table 4-4. Positions Monitored by the PFTIR in the Flare Plume, Averaging Times, and Average Temperature Observed

Time Activity Start Time Stop Time Avg. Temp.

(°C) 17:13:17 Start Acquisition 17:25:33 Just above combustion zone (~1m) 17:25:33 17:28:19 302 17:28:36 Higher above combustion zone - cooler 17:28:36 17:31:07 293 17:33:33 Raise front leg of tripod 17:36:36 2-4 m above combustion zone 17:39:44 Start left extreme move right 17:41:55 Centerline 17:41:55 17:43:03 222 17:43:54 Right of center 17:47:16 Back to centerline 17:53:18 Flamelets getting into FOV 17:56:18 Lower just above flame zone (~0.5m) 17:55:46 17:57:39 412 17:57:39 Left of center just above flame 17:59:37 Right of center just above flame 18:02:11 Shut off flame

4-14

Deviations from the Test Plan – Controlled flare tests proceeded essentially as anticipated. Both vertical and horizontal traverses and centerline data were collected at various heights above the combustion zone as discussed above. A variation from anticipated testing was in the measurements in which the FOV was traversed to different locations in the plume. No clearly defined plume gradient was observed. Instead a statistical ensemble of warmer and cooler turbulent parcels of gas was found to be passing through the FOV of the PFTIR. These represented the inhomogeneous mixing of the exhaust with the ambient air. While measurements were made at different heights and at different horizontal positions from the “centerline” of the combustion zone, a distinct gradient could not be seen. 4.2 Analytical Procedures The analytical procedures that were used to determine the concentrations of target species are described below for EFTIR, canister, and PFTIR samples. 4.2.1 Extractive FTIR – Plume Generator Tests

FTIR spectroscopy exploits the infrared (IR) region of the electromagnetic spectrum for qualitative and quantitative analysis of chemical compounds in gas samples (in this case, from stacks). In EFTIR, a flow sample of the gas stream being tested is drawn from the stack and flows through the sample cell of the instrument.

The IR light beam is generated by a ceramic element, glow bar source that is heated by passing electrical current through the element. When heated, the element emits light in the IR portion of the electromagnetic spectrum. The light is then collimated (i.e., focused in a beam) and is modulated with an interferometer. When the IR radiation interacts with molecules, the molecules absorb portions of the radiation causing the molecules to rotate or vibrate. The precise frequency of the radiation that the molecule absorbs, causing these rotations and vibrations, depends on the functional group in the molecule. The IR beam passing through the gas mixture for defined path length is read by an optical detector, and after signal conversion, the IR absorbance spectrum is recorded.

Molecules that absorb IR radiation show a characteristic pattern of absorbed frequencies called an absorbance spectrum. The absorbance spectrum is unique to each individual molecule, like a fingerprint, that can be used to identify and quantify a particular species in a complex mixture of gases. As governed by Beer’s Law, the magnitude of a compound’s IR absorbance is directly proportional to the product of its concentration in the mixture and the optical path length.

4-15

The optical path to which the stack gas is exposed will be partially absorbed by various molecules present and captured by the IR-sensitive detector to produce a spectrum that represents the proportion of absorbed light spread over the entire near-IR spectral range (400 cm-1 to 4500 cm-1).

The quantitative analysis is performed by matching the features of an observed spectrum to reference standards in a spectrum library to identify the presence of a given compound. The quantity of compound is computed automatically by a software program in the computer associated with the device. Results are presented in a numerical data format, usually in spreadsheet form. The spectra and a time series of concentration for the test gas can both be created and displayed.

For the current application, the list of standards included carbon dioxide (CO2), carbon monoxide (CO), butane, ethylene, propylene, and propane. When more than one feature was present in the same region, a linear combination of references was used to match the compound feature. The standards were scaled to match the observed band intensities. This scaling also matches the unknown compound concentrations. The scaled references were added together to produce a composite, which represents the best match with the sample. A classical least squares mathematical technique was used to match the standards’ absorption profiles with those of the observed spectrum in specified spectral analysis regions. The regions were chosen carefully to provide optimum detection of the compounds of interest with minimum interference from other compounds. Compounds of interest and any known compounds expected to present spectral interference (water [H2O] and CO2 in this case) were included in the reference set of compounds.

The amount of radiation that is absorbed by a compound is directly proportional to the amount of the compound present (or the compound’s concentration). Therefore, by measuring the amount of radiation that is absorbed when IR light passes through a mixture of compounds, the concentration of each compound was determined.

As shown in Figure 4-7, the IR light beam makes multiple passes through the sample chamber where molecules absorb frequencies characteristic to the individual compounds. The spectral data were processed in real-time to yield compound identities and concentrations. For the MKS FTIR instrument used in this test program, the IR beam was reflected multiple times by a set of optically matched gold-plated glass mirrors. The mirrors were gold plated to minimize

4-16

Figure 4-7. Schematic Diagram of EFTIR Apparatus Measurement

degradation of the reflective surface from moisture or any other reactive gases in the exhaust stream. The cell path length was fixed at 20:1. After exiting the cell, the IR beam was directed into a liquid-nitrogen cooled mercury/cadmium/telluride (MCT) detector, a photoconductive device that produces an electrical voltage proportional to the amount of IR light that strikes it. The EFTIR sample cell and the extraction lines were maintained at a temperature of 160°C to prevent condensation losses and minimize interactions with the Teflon line. Instrumental resolutions were at 0.5 cm-1, and signal averaging was performed for 40-second periods.

The spectrum analytical method used for testing at the John Zink facility was developed by selecting the spectral regions and subregions that are least affected by the primary IR absorbers, H2O and CO2, while also producing the best detection limit possible for the target compounds. Typically, an analytical method will be iteratively refined by using it to analyze a representative set of IR spectra while varying the method. The optimum method is indicated when both the 95 percent confidence levels and the bias on the individual compounds are minimized. Table 4-5 summarizes parameters of the references used in the analytical method, with respect to optical depth and analysis region, for each compound monitored.

Detector

Modulated (FTIR)

IR Source

IR Multi-pass Sample Cell

Gas sample inlet

Gas sample outlet

Voltage signal output to spectral analyzer

4-17

Table 4-5. EFTIR Analytical Method Parameters for Target Compounds and Spectroscopic Interferants

Of the compounds listed in Table 4-5, all but the ambient species (H2O, CO2, CO, and

nitrous oxide [N2O]) experienced periods where their concentration fell below the detection limit

of the EFTIR. Minimum detection limits (MDLs) were calculated for those compounds

observed near the instrument’s noise level.

This was achieved by analyzing sets of data where only interferants were present (the compounds of interest were determined to be absent by spectral investigation). When applied to this series, the analytical method should return a set of values centered about zero. The MDL is then estimated by taking three times the standard deviation of that data scatter. If the data are scattered about a non-zero value, it is indicative of a method bias and bias correction procedures are invoked. Table 4-6 summarizes the MDL estimates for the EFTIR that were obtained for each target compound during the test program

Compound Reference Optical Depth of (ppm × m) Analysis Region

(wave number, cm-1) H2O 28,870 - 27,509,700 1097.31-1168.67

CO2 20,061 – 1,025,100 2211.02-2236.33

CH4 87.4 – 39,020 2838.98-2921.67

N2O 102 – 4,076 2124.96-2236.33

CO 25.7 – 20,358 2163.77-2184.50

Ethylene 85.9 – 2,576 900.12-1000.17

Propylene 2,030 815.76-1078.04

n-Butane 8,080 883.98-1064.54

Propane 169.7 – 3,468 2700.38-3081.50

4-18

Table 4-6. EFTIR MDLs for Target Compounds

a Detection limits for ambient air constituents were not calculated since they were always present.

4.2.1.1 Calibration and Quality Control Checks A series of system quality control (QC) checks and calibration procedures were conducted for the EFTIR measurement system to ensure proper system operations and data quality.

Reference Standards – An IR spectrum analysis was performed by matching the

features of an observed spectrum to those of reference standards. If more than one compound was present in the analytical region, a linear combination was used to match the features of each compound. The standards were scaled to best match the observed band intensities in the sample. Reference standards were available at 150oC for all target compounds except for n-butane, propylene, and propane; therefore, references for n-butane and propylene had to be generated prior to testing. This was performed by directly injecting the certified (± 2 percent) gas standard at a constant rate. Each gas standard was allowed to flow for approximately 15 minutes and then a 5-minute averaged (537 individual scans) spectrum was recorded. The gas standard continued to flow for an additional 10 minutes and a second 5-minute averaged spectrum was collected. The two spectra were compared to ensure that the cell and injection system were sufficiently purged. For propane, an available reference standard at 185oC was used. Even though the software accounts for the difference in reference temperature when calculating the concentration, significant error could occur if the absorbance profile differs greatly. Therefore, the propane

Compound MDL (ppmv)

H2O NA a

CO2 NA a

CH4 NA a

CO NA a

N2O NA a

Ethylene 0.28

Propylene 2.6

n-Butane 6.2

Propane 1.7

4-19

reference absorption profile was compare to the actual sample to ensure that the temperature difference had no effect on the band shape in the analysis region.

EPA Method 320 Spiking – Field validation of select target compounds, at various concentration levels, was conducted while on site. The validations were conducted using certified (±2 percent) standards spiked on top of stack gas. The quality assurance (QA) gas standards used are listed in Table 4-7.

Table 4-7. EFTIR Calibration Gas Standards Certified Gas

Standard Certified Constituent

Concentrations (ppmv) Tolerance CC88958 Ethylene 100

R-114 20

Nitrogen balance

+ 2%

SG1697NB n-Butane 402

R-116 20

Nitrogen balance

+ 2%

CC55106 Propylene 101

R-116 20

Nitrogen balance

+ 2%

The QA validations were conducted to characterize the extraction apparatus and cell interactions toward the primary compounds of interest and verify that the analytical method would return accurate concentrations. In accordance with EPA Method 320 and the project QAPP, QA spiking for the target compounds was performed to ensure that the FTIR system was operating correctly and producing data of the highest integrity.

Deviations from Method 320 spiking procedures were minor and were as follows:

• Pre-test calculations, with the exception of instrument operating calculations, were not necessary. Since the sampled gas stream was simulated, the expected concentration ranges were known.

• The Teflon™ calibration line that delivered the QA standards to the inlet spiking "T" was unheated. This line was virgin Teflon™ and had been purged with nitrogen to ensure that no water was present. These precautions ensured that each analyte was delivered to the spiking tee dry and contamination free.

4-20

• Since time and injection gas availability were of concern, QA spiking was done daily instead of before and after each test as recommended by EPA Method 320.

• To conserve injection gas, the QA spiking was not performed on top of the same sample matrix as dictated by Method 320. However, the spikes were performed at concentration levels similar to those expected and observed in the sample gas during the tests.

• The gas bottle regulator used on the QA standards was not a purgeable regulator.

Per EPA Method 320, QA spikes of target analytes were performed at three different

levels. In addition to the analyte of interest, each QA spiking gas standard contained a component to be used as a spectroscopic tracer (ST). Common properties to all spectroscopic tracers are that they exhibit a broad absorption profile over a large concentration range and they are chemically inert. The linear behavior of the spectroscopic tracer allows a precise measurement of the dilution ratio of the spiked gas to native gas. For the Method 320 QA spike, this dilution ratio was determined using Freon-116 or Freon-14 as the STs and applied to calculate the theoretical analyte concentration using the following equation (4-1):

AnalyteTheoretical = ( ) ( )stackcylinder

samplecylinder

cylinder

sample AnalyteSTSTAnalyte

STST

⎥⎦

⎤⎢⎣

⎡⎟⎠⎞

⎜⎝⎛−+⎟

⎠⎞

⎜⎝⎛ 1 (4-1)

where:

AnalyteTheoretical = Theoretical Analyte concentration (ppm) STsample = ST concentration (ppm) as seen by the FTIR during QA spiking STcylinder = Concentration (ppm) of ST in the certified gas standard Analytecylinder = Concentration (ppm) of the target analyte in the certified gas standard Analytestack = Concentration (ppm) of target analyte in the stack during stable conditions

The criterion for a successful QA test is a measured concentration within 0.7-1.3 times

the calculated theoretical concentration.

The EFTIR was calibrated against a certified gas mixture containing the compounds of interest in the carrier gas matrix at concentrations equivalent to or near those to be measured in the tests. This mixture contained CO, CO2, H2O as well as VOCs to simulate a flue gas stream. This gas mixture was installed with the EFTIR at the test site and was intermittently fed to the system to check instrument calibration between data collection samples, as needed, during the actual test period. The EFTIR operator verified that the concentrations, calculated from EFTIR

4-21

raw data analysis of the certified gas spikes, corresponded approximately to the correct species and concentrations. The calibration procedure used corresponds to that of EPA Method 320 for EFTIR systems.

Pre-Test Checks – After the EFTIR instruments were set-up on site, a series of initial diagnostic tests and calibrations were performed to demonstrate that each system maintained its integrity during shipment. These checks consisted of the following:

• Cell Leak Checks: This test checks the integrity of each cell by pulling a vacuum and then monitoring the leak rate. The acceptance criterion for this test is a leak rate of 2 torr/minute.

• Sample Cell Exchange Rate: Sufficient sample exchange rate in the IR cell relative to the spectral averaging period.

• IR Detector Linearity Checks: For best results, it must be assured that the IR detector yields a linear response throughout a reasonable absorbance range at all the frequencies of interest. A software linearizer is used to continuously adjust the MCT detector preamp signal in order to achieve the desired response. To optimize the linearizer, background spectra are acquired with and without a polyethylene film in the IR beam. Comparison of the strongly absorbing polyethylene bands in the low, mid, and high frequency regions against a clean background enables the processor to calculate and appropriately set the detector’s linearizer terms (offset, linear, quad, cubic and delay).

• Noise Equivalent Absorbance (NEA) or Signal-to-Noise Ratio (SNR) Tests: Signal-to-noise ratio (SNR) tests were performed by measuring the noise equivalent absorbance (NEA) of each FTIR system. A NEA spectrum was obtained by collecting two consecutive single beam spectra, while the sample cell was being purged with UHP nitrogen, then forming a ratio of them to one another. This effectively produces a noise or “zero” spectrum in which the peak-to-peak absorbance is determined. This provides a measure of the sensitivity of the instrument for the specified spectral resolution and number of scans.

Periodic Checks – A series of on-site validations and system checks were performed

on the EFTIR and sampling system to ensure data of known quality. The following system checks were performed daily:

• Spectral Background: A spectral background is essentially a “blank spectrum” in

that it does not contain any of the target compounds present in the sample. It was created by purging the cell with ultra-high-purity (UHP) nitrogen while collecting a spectrum. The analytical software then formed a ratio with this spectrum against each

4-22

collected sample spectrum to produce an absorbance spectrum for quantitative analysis.

• Spectrometer Frequency and Resolution Checks: A real-time check of frequency position and resolution was performed each day by monitoring a specific water absorption line (present in ambient air). The position of this line must not deviate more than ±0.005 cm-1 from the reference value over the course of each test. Likewise, the line width (directly related to instrument resolution) must not deviate more than ±0.05 cm-1 from the reference value over the course of each test.

4.2.2 Canisters

Canister samples were collected and analyzed for both the plume generator and the controlled flare tests. The purpose of the plume generator canister samples was to verify the concentrations of the spiked gases that were measured by the EFTIR. Canister samples of the fuel being fired by the controlled flare were collected to confirm the composition provided by the gas vendor. Since the constituents of interest were different between the two, the analyses also differed somewhat. The analytical procedures for both are described in the following sections. 4.2.2.1 Plume Generator Tests

Plume generator stack gas canister samples were air shipped to ATL for analysis according to ASTM Method D-1945 for the following constituents: CO2, CO, oxygen, butane, and total nonmethane organic compounds (NMOC). In addition, ATL Method @82 was used to quantify ethylene and propylene concentrations. ASTM 1945 employs either gas chromatography of flame ionization detection (GC/FID) or GE/TCD to quantify gas constituent with direct injection of up to 1 mL of sample. ATL @82, an in-house modified version of TO-14a developed by ATL, uses direct injection of a 1.0 mL sample through a fixed loop followed by quantification by GC/FID. Canisters were pressurized with nitrogen in the lab prior to analysis. Additional details regarding sample analyses are provided in the laboratory analytical reports provided in Appendix F. 4.2.2.2 Controlled Flare Tests

Propane fuel canister samples were air shipped to ATL for analysis according to ASTM Method D-1945 for the following constituents: CO2, CO, oxygen, methane, ethane, propane, isobutane, butane, neopentane, isopentane, pentane, and C6+ hydrocarbons. This method employs either GC/FID or GE/TCD to quantify gas constituents. Canisters were pressurized with nitrogen prior to analysis. Because these samples were expected to contain over 90 percent propane, well above the upper calibration range of the instrument, propane analyses were

4-23

conducted after additional dilution of the sample gas to obtain a concentration within the calibration range of the instrument. All other analyses were conducted after initial pressurization of the canisters. Additional details regarding samples analyses are provided in the laboratory analytical reports located in Appendix F. 4.2.3 Passive FTIR The fundamental principles of PFTIR measurement, as described below, apply to both the plume generator tests and the controlled flare tests. Differences occur only in details of device location relative to the source being measured, fields of view during sampling as dictated by distance from the source, and data treatment based on the differences in temperature between the test plumes. Two data treatment and analysis techniques were used to analyze the PFTIR data collected: 1) a “near real time” technique using Classical Least Squares (CLS) procedures[1] and 2) a more rigorous “post test” technique using line-by-line (LBL) techniques.[2] Each of these techniques are also described in this section. 4.2.3.1 Principles of Passive FTIR Measurement

PFTIR operates on the principle of analyzing the thermal radiation emitted by hot gases. In normal absorption spectroscopy, light is passed through a region containing a gas to be analyzed, and the transmitted light is captured and then spread out into a spectrum using an interferometer (FTIR) or a spectrometer. In this spectrum, the presence of specific compounds can be determined from the patterns of the light absorbed, while the concentrations of the compounds can be measured from the intensity of these patterns. The low energy of IR light is absorbed by molecular species causing stimulation of the various vibrational and rotational modes within the molecule. Because each molecule consists of a unique structure of bound atoms, the patterns of IR wavelengths (IR colors) absorbed are also unique. These molecular “fingerprints” are used to uniquely identify and quantify gases and their concentrations.

To monitor flares, standard absorption IR spectroscopy could be used. However, it is logistically difficult to pass an IR light beam through an elevated flare plume and then capture the transmitted light. A passive approach is an attractive alternative. Passive means that no “active” IR light source is used. Instead, the hot gases of the flare itself become the IR source. The spectrometer is now simply a receiver. This approach is possible because the IR radiation emitted by hot gases has the same patterns or “fingerprints” as their absorption spectrum. Consequently, observing a flare with an IR instrument allows for identification and quantification of species through emission spectroscopy just as can be done with absorption

4-24

spectroscopy. There is one main difference between these two approaches: the emission signature arising from a hot gas is proportional to the concentration of the gas as in absorption but it is also affected by the gases’ temperature. Therefore, in conducting emission or radiance measurements, the temperature must be deduced in addition to the gas concentrations. Consequently, unlike absorption spectroscopy, the PFTIR signal must be calibrated in absolute units of radiance. This requires that the instrument be calibrated utilizing a light source of known spectral radiance.

The radiance calibration of a PFTIR instrument relates the output voltage of the instrument to the received energy in radiance units. Radiance is given by watts per square centimeter of source area, per unit solid angle of observation, and per unit wavelength or wave number of detection (watts/cm2/steradian/wave number). To do this calibration, one uses an infrared blackbody source. An IR blackbody is simply an object that is perfectly absorbing throughout the IR. Any object absorbs and emits radiation in the same proportion. Consequently, if an object is perfectly absorbing, it is also perfectly emitting. The Planck radiation law gives the power emitted by a perfect black body as a function of temperature, T, and wave number, ν. This function is given by:

( )1

2, /

32

kThcvBB evhcTvN = (4-2)

Here, h is Plancks constant, C the speed of light, and k Boltzman’s constant. This function looks like that shown in Figure 4-8 at temperatures of 200, 500, and 700°C. As the temperature rises, the peak of this function is seen to move toward larger wave numbers (smaller wavelength) and its intensity increases. For this reason, hotter objects will emit more in the short wavelengths or visible regions while cooler bodies emit in the longer wavelengths or the IR regions.

If a body is not “black” or totally absorbing, the energy it emits is the Planck function multiplied by the body’s emissivity which is in most cases just its absorptivity. For example, if the body is 50 percent absorbing, it will emit 50 percent of the Planck function. In the case of gases, they have highly variable absorption with wavelength. In fact, it is this variation that produces the absorption patterns that allow for the identification of gases in the IR. If the

4-25

Figure 4-8. Planck Function Showing the Radiation Emitted by a Blackbody at 200°C

transmission of a gas is given by τ (ν,T) then [1 - τ (ν,T)] is the amount of absorption. The radiation emitted by this gas at temperature T and wave number ν is then given by:

N (ν, Τ) = [1 - τ (ν,T)] * Nbb (ν,T) (4-3)

For flare measurements, it is the signal of Equation (4-3) that is being detected from the hot gases above the combustion zone. However, there are also other contributions to the signal that an analyzer “sees” or detects in the field. As shown in Figure 4-9, the background (typically the sky) has some IR emission; this is also defined by Equation (4-3). When transmitted through the plume and the intervening atmospheric path between the plume and the PFTIR it is detected and becomes part of the overall signal. The plume emissions given by equation 4-3 are also transmitted through the atmospheric path between the flare and the PFTIR. This constitutes the signal of interest. To complicate things more, the atmosphere between the flare and the PFTIR also has its own IR emissions. The total radiant signal received then consists of:

0.0E+00

1.0E-04

2.0E-04

3.0E-04

4.0E-04

5.0E-04

6.0E-04

0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000

Wavenumbers (cm-1)

Rad

ianc

e (w

/cm

2/st

r/cm

-1)

200C

500C

700C

4-26

Background Radiance Flare

Radiance

AtmosphericTransmission& Radiance

Figure 4-9. Contributions to Flare Radiance

4-27

Ntotal = Nbkg * τflr * τatm + Nflr * τatm + Natm (4-4)

where the arguments ν,T have been dropped for clarity. In this equation, Ntotal = total radiance Nbkg = background (sky) radiance τflr = flare gas transmission τatm= atmospheric gas transmission Nflr = flare gas radiance Natm= atmospheric gas radiance In some cases, such as a very high temperature flare or in regions of the spectrum where the absorption is high (e.g. the CO2 bands), the background radiance and the atmospheric path radiance can be negligible. If this is true, Equation (4-4) reduces to just the plume radiance multiplied by the atmospheric path transmissivity. If any of these terms are not negligible, they must be accounted for in the analysis algorthim. However, the PFTIR provides the capability of obtaining the necessary atmospheric and background radiance data. Specifically, the PFTIR provides: 1. The atmospheric path transmission over the horizontal observation path; 2. The background radiance as seen through the slant atmospheric path; and 3. The total radiance as defined by Equation 4-4. Atmospheric Path Transmission – The atmospheric path transmission is measured over a horizontal path and this spectrum is analyzed to determine the concentrations of gases in the local ambient air. These concentrations are then used to compute the slant path transmission between the flare and the instrument using line-by-line (LBL) codes[3]. This process then gives the τatm term in Equation 4-4. If τatm is known, the atmospheric path radiance can be computed directly using Equation 4-3. If necessary, multiple atmospheric layers can be included in the calculations to match the atmospheric temperature and pressure profiles as predicted by standard atmospheric models. In general, this is not required unless the flare is exceptionally tall. Background Radiance – The background radiance is measured by setting the PFTIR field of view to the side of the flare and looking at the sky with the same observation angle as used in the flare measurement. This then gives the term:

4-28

τatmbkgN * (4-5)

which is the same as the first term in Equation 4-4 except for the plume transmission. In general the sky radiance is variable exhibiting emissions from high altitude gases (e.g. O3) when the sky is clear while looking a lot like a pure Planck function at 300K when the sky is cloudy. Plume Transmission and Radiance – Once the background and atmospheric path transmittance are measured, the only unknowns in Equation (4-4) are the plume transmission and the plume radiance. These are again related by Equation (4-3). The process of measuring the plume concentrations then reduces to determining plume temperature and plume transmission. This is done using LBL codes in an iterative calculation to match all spectral features in the measured data or by approximate CLS codes using libraries of spectra at the appropriate temperature. These two approaches are discussed below. 4.2.3.2 Data Analysis Techniques There are two analysis techniques that were employed to interpret the data obtained by the PFTIR: Classic Least Squares (CLS) and Line-by-Line (LBL). The CLS analysis may be performed much more quickly, and provides a near real-time result, while the LBL analysis is more complex, and requires significantly more time to arrive at a solution. The LBL analysis is also considered to be the more rigorous of the two approaches, and so the results presented in Section 5 were developed using LBL. (Appendix C contains results of the CLS analysis, and in some cases, the CLS produced values that agreed better with the EFTIR. The evaluation of the suitability of the two analysis techniques continues at the writing of this report, and is anticipated to be part of the next phase of the PFTIR method validation.) Post-test Line-by-Line Reduction Algorithms – The LBL approach can deduce temperature as well as a gas concentrations because the shapes of the bands change with temperature and their intensity with concentration. The change in band shape with temperature is shown in Figure 4-10 for carbon monoxide. Here, the top plot is CO at 300K (27°C), while the lower plot is CO at 550K (277°C). The dominant effect of increasing temperature is to

4-29

CO 50 ppm 300K 760 Torr 1meter path

-0.000

0.005

0.010

0.015

0.020

0.025

0.030

0.035

Abs

CO 50 ppm 550K 760 Torr 1meter path

0.000

0.002

0.004

0.006

0.008

0.010

0.012

0.014

0.016

Abs

2000 2050 2100 2150 2200 2250Wavenumbers (cm-1)

Figure 4-10. Structure of the Fundamental CO Band at 300K (top) and 550K

(bottom) Showing Alteration of Band with Temperature expand the band (increasing the strength of the weaker lines farther from band center) while shifting the peak position away from band center. This changing of band shape combined with the overall increase in radiance due to the multiplicative and highly temperature dependent Planck function (Eq. 4-3) makes the plume spectral radiance level and shape highly dependent on temperature. An iterative calculation matching both band shapes and intensities for all compounds present is essentially a measure of temperature and gas concentration simultaneously. The starting point of the iterative calculations is equation 4-4. By iterating on the gas concentrations, temperature (for the flare test), and the strength of the background radiance it was possible to obtain fits of the theoretical model to the data. The absorption and emission

4-30

spectra for various gases within the plume and the atmosphere were calculated utilizing a specially modified version of the E-Trans line-by-line computer program and a modified version of the HITRAN database. The iterative model also included a modified version of the Non-Lin open-path data reduction code, which accounts for non-linearity in absorbance with increased gas concentration and/or low spectral resolution. In typical open-path atmospheric applications these effects are relatively small at low absorbance values or in data acquired at high spectral resolution. However, this was included due to the high concentrations of CO2 and the narrower line widths of the spectral lines due to the elevated plume temperatures. The spectra used for the hydrocarbons (butane, ethylene, propylene, and propane) in the analysis were the highest temperature reference spectra available. All computations were density corrected using pressure and temperature in the ideal gas law. However, if the temperature of these references were significantly different than the measurement conditions a mismatch in shape would persist. The LBL calculation produces a spectrum by combining the effects of each spectral line for each compound specified in the calculation. In the CO plot of Figure 4-8, for example, there are more than 50 CO lines in the upper plot at 300K and 68 to 70 in the lower plot at 550K. In a full calculation, hundreds of thousands of lines may be involved. A specially modified version of the HITRAN database was used in this calculation. The original database was developed to treat atmospheric transmission; therefore, the data have been extensively tested for normal atmospheric constituents like H2O, CO, CO2, CH4, N2O, NO, etc. at ambient temperatures However, this database lacks much of the data required for modeling high temperature phenomena. An augmented version was therefore utilized with the E-Trans computer program to correctly model the high temperature spectra of the flare. The HITRAN database contains the strength of each spectral line, its pressure broadened width, its lower state energy (the lower molecular energy level involved in the particular transition producing the line), the quantum numbers of the transition involved, and an isotopic identification for the compound. These parameters allow the full absorption or transmission spectrum of gases to be computed. The lower state energy is important because as temperature changes, the number of molecules in the various energy levels shift. At higher temperatures, more of the molecules have higher energy and therefore, they move up to higher energy levels. At lower temperatures, more of the molecules are in the lower levels. This effect produces the band shape change shown in Figure 4-10. The Boltzmann distribution dictates the relative population of each state of energy E”. This distribution is given by:

4-31

( )

" /E kTonn eQ T

−= (4-6)

where: n is the number of molecules in energy level E”, no the total number of molecules in the sample k is Boltzmann’s constant, T the absolute temperature, and Q(T) is a constant for each molecule called the partition sum. It is essentially the sum

of the exponential in the numerator over all possible energy levels of the molecule.

This function shows that as T increases, the number of molecules in a particular state will increase or decrease depending on whether E” is larger or smaller than kT. This distribution is used to determine the population in every transition (spectral line) computed so that each line will have the proper strength for the temperature being considered. Given the populations, the line strengths are adjusted for temperature and the effect of all lines combined using:

e LPKΣΣ)T,(τ jiijT)(ννν −= (4-7)

where: Ki is the absorption coefficient of the i-th spectral line, Pj is the partial pressure of gas j, and L is the path length. The sum over i is for all lines of a given species, while the sum over j is over all species in the calculation. In this case, the Ki factors identify the spectral line shape for the lines. Typically, this is the Voigt profile, which is a combination of line broadening due to collisions (pressure broadening) and broadening caused by Doppler shifts arising from the rapid motion of the molecules. Typically, near atmospheric pressure, the Doppler influence is very small, and the line is essentially pressure broadened. However, at elevated temperatures, Doppler effects can become more significant because the velocities of the molecules are increased. Therefore, the Voigt profile was used in all calculations for the plume study. In practice, the atmospheric path from the plume to the PFTIR contains some pressure and temperature gradients. This is also true for the plume itself, although the measurements performed in this study were taken well above the combustion zone so they are not as severe as

4-32

measurements taken near the flame. To treat gradients, the path is divided into small zones each of which can be considered isothermal and isobaric. A separate calculation as in Equation 4-7 above is then performed over each zone with the final result being the combination of all zones. When measuring a plume exhaust, it is necessary to match any gradient that exists in the plume. The severity of these gradients were assessed by making measurements across the plume from near ambient condition on one side to near ambient conditions on the other.

Near-real Time CLS Reduction Techniques – Classical Least Square analysis is a technique developed to fit measured absorption spectra and deduce gas concentrations[1]. The fundamental measurement in absorption spectroscopy is transmission. Because the problem of measuring flare gas concentrations reduces to that of measuring plume transmittance, the same CLS techniques can be applied. The restriction imposed by CLS is that the gas mixture being observed is assumed to be a single isothermal and isobaric volume. Unlike the LBL technique that accounts for thermal and concentration gradients, current CLS approaches cannot treat these conditions. If gradients are significant, the CLS approach will produce higher than actual efficiencies (as shown from the data presented in Appendix C of this report). However, the primary advantage of CLS is its speed. Analysis of a full IR spectrum by CLS requires a few hundred milliseconds. In contrast to this, LBL analyses require several minutes to tens of minutes per spectrum depending on the complexity of the scenario. This makes CLS a good choice for approximate real-time analysis during data collection. If the temperature and pressure gradients are not a major issue, CLS and LBL results will theoretically converge to the same result. It is yet to be determined if this will be the case for typical industrial flares. The transmittance of a gas is given by

Transmittance = ( ) cl)v(kev −=τ (4-8)

Here κ(ν) is the absorption coefficient of the gas, c the concentration, and l the path length occupied by the gas. κ(ν) is essentially a constant at a given temperature but, as discussed above for LBL calculations, it does vary if temperature changes significantly (>50 to 100°C). κ(ν) is essentially the absorption standard which is contained in the reference libraries of FTIR systems.

Because equation 4-8 is exponential and very non-linear, a quantity called absorbance is defined which is the log of this equation. Absorbance is then:

4-33

Absorbance = Abs(v) = -log10 τ(ν) = (0.434) κ(ν) cl (4-9) Log base 10 is used for historical reasons and 0.434 is just the log base 10 of e. Equation 4-9 is linear in concentration multiplied by path length. This allows it to be more easily fit to collected data. For a real world plume, more than one compound is always present. In the gas phase, the transmittance of a mixture is just the product of the individual gas transmittances. This is equivalent to taking the sum of all gases in the exponential of Equation (4-8). The absorbance of a mixture of gases therefore becomes:

( ) 1 ( )j jj

Abs v v cκ= ∑ (4-10)

To deduce gas concentrations from an observed absorbance spectrum, one needs to find the set of Cj concentrations that minimizes the error between the sum of Equation 4-10 and the observed absorbance spectrum, at all wave numbers, ν. This is essentially a least squares problem, hence the name Classical Least Squares or CLS. If Equation (4-10) is written in digital form specifying Ai as the absorbance at data point i, this becomes:

,1i i j jj

A Cκ= ∑ (4-11)

this can be put in matrix form as:

C~~1A~ •= κ (4-12)

The solution to this matrix equation is:

[ ] A~~~~11C~ T1T ••=

−κκκ (4-13)

Where T~κ is the transpose of the absorption coefficient matrix and [ ] 1T ~~ −• κκ is the inverse of

the product of Κ and its transpose. This is essentially the least squares fit of the series given by Equation (4-10) to the data given by Abs(ν). Classical least squares essentially provides the concentrations, c, which best reproduce the observed spectrum. Because the least squares

4-34

process is fitting to a large number of data points, a standard error can be determined from this process. This is essentially the spectral residual represented by:

[ ] δκ ~C~~1A~ =•− (4-14)

The residual in each analytical region is converted to a gas concentration for each

compound fit in that region using its reference spectrum. This provides a standard error for each gas concentration produced. However, the standard error only indicates how well the measured spectrum was reproduced by the fitting process. It is possible, particularly with radiance data, that this error could be very small yet a systematic error could still produce a significant effect. This could be the case, for example, if the sky radiance spectrum or the calibration curve used to get plume transmittance from the observed radiance spectrum were incorrect.

The CLS approach has to make certain assumptions about the test scenario when it is applied. As mentioned above, current CLS techniques solve for absorbance or equivalently transmission of one isothermal, isobaric path. This does not mean, in the present case, that the entire path from sky to PFTIR needs to be isothermal and isobaric but it does demand that the plume itself must be. A measurement of temperature is made for the CLS calculations using either the observed band envelope of CO lines or using ratios of closely spaced water lines that show a change in relative intensity with temperature. This gives an average temperature through the plume. The measured spectra (path transmission, sky background, and observed radiance) are combined to produce an absorbance spectrum of the plume. The CLS technique then fits the absorbance spectrum at the mean temperature and deduces individual gas concentrations. Like all CLS approaches, different spectral regions are used to fit the various species. Each region is chosen to provide the strongest feature of each analyte with the least interference from other compounds. However, interferences will always be present and the CLS routines take these into account in performing the fit.

The CLS routine used in this program also has built into it algorithms for linearizing the

absorbance for each analyte with concentration. This eliminates a major source of error in CLS, that of assuming linear behavior as anticipated by equation 4-9 over all concentrations and spectral resolutions. The output of the CLS algorithm is in units of concentration time path length or ppm*m. Since the exact dimension of the plume was not known this quantity was not converted to concentration. However, for efficiency calculations these units are adequate. In the CLS analysis, the measured horizontal path transmission was used directly assuming little

4-35

difference between it and the slant path transmission. For use in measuring tall flares, the CLS will need to solve the horizontal path transmission for gas concentrations and then compute a slant path transmission using appropriate atmospheric models for temperature and pressure fall off with altitude. 4.2.3.3 PFTIR Calibration

To perform passive measurements as in this program, the PFTIR must be calibrated as a radiometer. This consists of determining the spectral response of the instrument for given radiant energy input. To perform such a calibration, a source of known radiant intensity must be used and energy from it must fill the field of view of the receiver telescope. This was accomplished using a NIST traceable blackbody source collimated by a 12-inch telescope identical to that of the receiver so it would fill the receiver aperture. The blackbody source used was an Omega model BB704 with an emissivity of 0.95 and a maximum temperature of 398°C. The radiance from this source will follow Planck’s radiation law times its emissivity. Therefore, the radiance of the source is given by:

( )1eνch20.95T),(λL Tk/νch

32

bb−

= (4-15)

where c is the speed of light, h Planck’s constant, ν the wave number of the light, k Boltsmann’s constant, and T the absolute temperature. This function was shown in Figure 4-8 above at 200°C, 500°C and 700°C. As seen above, the intensity of the Planck function increases with temperature and its peak shifts toward the visible. At the low temperatures present in the flare program, the peak of the energy distribution is near 1000 cm-1 and the energy is essentially zero by 3500 cm-1. It is the low energy at 3000 cm-1 that makes detection of total organics difficult when the plume temperature is low.

The calibration of the PFTIR is accomplished by measuring the blackbody at several temperatures and taking the ratio of the Planck function divided by the measured signal. This gives a calibration in radiance units per volt of response at each wave number. An example of the measured radiance at various temperatures is given in Figure 4-11. In reality these plots are not used directly because of the atmospheric H2O and CO2 absorption in them. Instead, a polynomial fit to the baseline of these plots is created as shown in Figure 4-12. The features between 1000 cm-1 and 1500 cm-1 seen in these latter plots are features of the optics in the instrument and are real variations in response so they are preserved. The plots of Figure 4-10 are

4-36

Figure 4-11. Measured FTIR signal using the black body source at 150°C, 250°C.

Figure 4-12. Background Fits to the Spectra of Figure 4-11 Eliminating the Atmospheric Absorption Features

4-37

then ratioed to the Planck function to get the system response. Generally, each of these radiance plots should provide the same response function if the system is linear. That is the case here as shown in Figure 4-13, where the response functions obtained with 150°C, 250°C, and 316°C data are shown. The only significant difference in these response functions is at the short wavelength (large wave number) end of the spectrum where the signal is very low at low temperatures and higher at higher temperatures. It would be expected that the higher temperature measurements would be more reliable in this region. The response function actually used in data reduction, was the average of the two higher temperature measurements at 250°C and 316°C.

Figure 4-13. System Response Determined from 150°C, 250°C and 316°C Measurements

Note that the power of ten on the y-axis is 10-6.

4-38

The accuracy of the radiance calibration in terms of the temperature accuracy of the blackbody source is given by:

Tδ

eT

ν *1.43881

)T

ν(*1.4388

N

Nδ

⎥⎦⎤

⎢⎣⎡ −

−

−= (4-16)

where: N = radiance, ν = wave number, and T = absolute temperature, K. This means at the Planck peak with a temperature of 500K and an uncertainty of 0.8K,

the percent radiance uncertainty is 0.48 percent.

For all testing, the radiance calibration was performed over a temperature range from 100°C to 250°C (373 K to 523 K), which spanned the temperature range expected for all plumes. The calibrations include an instrument-background emission measurement. This was obtained by looking at a cold target (liquid nitrogen cooled aluminum) placed at the focal point of the collimator. The radiance calibrations were performed at the beginning and end of each test day to account for any possible variation in the instrument emission. These emissions change if the instrument temperature changes significantly during the measurement period.

Other Data Treatment Considerations – Data reduction for all tests essentially

followed the same steps. The radiance calibration function of the PFTIR was first deduced from the blackbody measurements at 150°C, 200°C, and 250°C. This was used to calibrate the measured data in radiance units (watts/cm2/str./cm-1). Sky radiance and atmospheric transmission measurements were then conducted. The data from the atmospheric transmission measurements were used to compute the slant-path transmission and radiance of the air between the instrument and the gas plume. The measured sky radiance was used along with the atmospheric path transmission to provide a first order estimate of sky effects. The interative LBL routine then deduced the plume transmission from these measurements which also provided individual gas concentration in the plume. In the case of the LBL fitting routine, the horizontal path atmospheric transmission was used to deduce individual atmospheric gas concentrations (e.g. CO, CO2, N2O, NO, etc) in the local air. These were then used in a separate LBL calculation to compute the slant path transmission from the PFTIR to the flare plume. If

4-39

necessary, atmospheric models were used to take into account the slight gradients in temperature and density which might exist from ground level to the plume altitude. In the CLS approach, as discussed above, the measured atmospheric path transmission was used directly

Some variation was observed in the instrumental radiance calibration from day to day.

Therefore, a baseline adjustment was used in the LBL code to account for the slight drifts occurring in the instrument calibration. The deviations in the radiance calibration came in the form of an offset of the calibration curve. This was compensated for in data reduction by including a variable offset in the radiance curve-fitting algorithm. The net result of this shift was a scaling of gas concentrations. Over regions such as that used for CO and CO2 determination, this scaling was essentially constant; therefore, all gases scaled the same. This means the absolute gas concentrations could be biased, but to first order the computed combustion efficiency, being a ratio of concentrations, would not be affected. Because there were only slight deviations in the various calibration curves, the laboratory calibration curve was utilized for the reduction of all data presented here. This was the case because the lab calibration was considered the best representation of the instrument’s response due to the time and care taken in making the measurements (compared to what was possible in the field operations). Any deviation from this curve would be compensated for by the fitting algorithm. In future studies, it will be necessary to determine why this calibration function is shifting since it this is not expected unless the instrument is changing temperature significantly, THE throughput of the system is changing due to contamination of optics, or possibly imperfect alignment of the collimator and the FTIR telescope during calibration.

The flare data was analyzed for temperature, and concentrations of CO2, CO, propylene,

ethylene, butane, and water. The region from 2050 to 2400 cm-1 was utilized for determining the temperature, and the concentrations of CO2, CO, and H2O. Although, H2O is of no practical importance, it appears as an interference that must be included in the fit procedure to obtain physically reasonable fitting results. Figure 4-14 depicts the spectral radiance data obtained from the flare test at time 17:14:26 on 27 August. The fit obtained from the LBL algorithms in the CO/CO2 analysis region is plotted with the data in Figure 4-15. Ethylene and propylene were determined by fitting the spectral radiance in the region of 880 to 980 cm-1. As before, water was also included in the fit of this region as it is an interfering species with spectral features that overlap with those of ethylene and propylene. The data and the resultant fit in this region are shown in Figure 4-16. Notice that in the data and fits the radiance levels drop below 0.0 at certain regions. These non-physical negative radiance values are an artifact of the shifting of the

4-40

-5.0E-06

0.0E+00

5.0E-06

1.0E-05

1.5E-05

2.0E-05

2.5E-05

3.0E-05

2050 2100 2150 2200 2250 2300 2350 2400

Wavenumbers

Rad

ianc

e [ W

atts

/(cm

**2

str c

m**

-1)]

Radiance Data Radiance Fit

Figure 4-15. Flare Test Radiance Data and Radiance Fit in the CO2 and CO Emission Spectral Regions

-1.0E-05

-5.0E-06

0.0E+00

5.0E-06

1.0E-05

1.5E-05

2.0E-05

2.5E-05

3.0E-05

3.5E-05

4.0E-05

500 1000 1500 2000 2500 3000 3500 4000

Wavenumbers

Rad

ianc

e [ W

atts

/(cm

**2

str c

m**

-1)]

Figure 4-14. Spectral Radiance Obtained on the Flare Test at 17:14:26 on 27 August 2003

4-41

-5.0E-06

0.0E+00

5.0E-06

1.0E-05

1.5E-05

2.0E-05

2.5E-05

3.0E-05

880 900 920 940 960 980

Wavenumbers

Rad

ianc

e [ W

atts

/(cm

**2

str c

m**

-1)]


Figure 4-16. Flare Test Radiance Data and Radiance Fit in the Ethylene and Propylene Emission Spectral Region

-6.0E-07

-3.0E-07

0.0E+00

3.0E-07

6.0E-07

9.0E-07

1.2E-06

1.5E-06

2800 2850 2900 2950 3000 3050 3100

Wavenumbers

Rad

ianc

e [ W

atts

/(cm

**2

str c

m**

-1)]


Figure 4-17. Flare Test Radiance Data and Radiance Fit in the Butane Emission Spectral Region

4-42

calibration offset curve. Butane was fit in the region from 2800 to 3100 cm-1. This portion of the spectrum, shown in Figure 4-17, has very low signal and the resultant butane concentrations are probably questionable. Most of the features in this region are water emission lines. Incorporation of a detector with high response, such as an InSb detector, would improve the signal-to-noise ratio in this region and allow for better extraction of not only butane but all heavy organics.

4.2.3.4 PFTIR Validation and Quality Control

PFTIR data were validated by comparison with EFTIR and canister sample results and the metered gas flow rates for the plume generator tests. The methods by which PFTIR conforms to TCEQ quality control requirements are specified in Table 4-8. It is important to note that PFTIR applied to measurement of emissions from flares is a developmental technique; therefore, strict adherence to TCEQ quality control requirements was not always possible.

The largest uncertainty in the PFTIR technique is in the inversion process. This process must measure temperature as well as gas concentrations using the three measurements discussed above. These measurements are:

1. The measured sky background as seen through the atmospheric path to the plume;

2. The atmospheric path transmission as measured over a horizontal path and converted to a slant path; and

3 The total scene radiance as seen through the atmosphere from PFTIR to the flare.

In addition, an assumption must be made as to the spatial distribution of temperature within the plume. The analytical study investigated the effect of various profiles on measured spectral radiance values. This study (refer to Section 5.1) indicated that the gradients had only a moderate effect on the efficiency calculation even if it was totally ignored as in the CLS case. However, the study did indicate that if the gradients are significant, the computed combustion efficiencies will be greater than the true values. In this program, the plume generator measurements are themselves the validation of the CLS and LBL procedures.

For tall flares, there is sufficient pressure and temperature change with altitude. This means that the slant-path transmission from the PFTIR to the flare will not be the same as the horizontal path transmission (measured to the base of the flare). As discussed above, these

4-43

Table 4-8. PFTIR Conformance to TCEQ Quality Control Requirements Requirement Method of Conformance

Define the system’s working range using a multipoint calibration curve for each target compound.

The working range was determined by the tests being performed in this study. Estimates were generated in a laboratory analytical study and demonstrated in the field tests.

Demonstrate contamination potential from sampling and preparation procedures using media blanks and equipment blanks.

Before and after each spiking test, measurements were made on hot ambient air blanks drawn through the plume generator. For the controlled flare tests, ambient air blanks were obtained by aiming the PFTIR to the side and upwind of the flare.

Demonstrate analytical equipment contamination potential using instrument blanks immediately after the highest concentration in calibration curve.

See above.

Demonstrate calibration bias and measurement precision by analyzing a second source standard four times (minimum).

Each test consisted of 15-20 individual spectra. These were independently reduced allowing for a standard deviation of the results to be computed and compared to the analysis of a “mean” spectrum. Radiance calibrations was performed each day of testing to correct for any calibration drifts.

Demonstrate calibration accuracy by analyzing a second source standard.

The accuracy was based on the comparison between the measured spectra, the known spiking compounds, and the standard spectrum for each compound in the spectral library. The PFTIR results were compared with computed concentrations based on metered flows and EFTIR measurements in the plume generator tests.

Determine method detection limits (MDLs) according to 40 Code of Federal Regulations (CFR) Part 136, Appendix B (i.e., measuring variability of seven replicate spikes), and then test these MDLs by analyzing a low level standard.

Data were gathered on heated air blank samples in the plume generator at the same temperature as typical simulations. Because the concentrations in this flow were low and constant (as constant as the ambient air levels), these spectra were used to determine a standard deviation of the gas concentrations deduced from the analysis procedures. Three times the standard deviation of these concentrations were then used as the MDL level.

4-44

Table 4-8. (continued) Requirement Method of Conformance

Evaluate collection efficiency, maximum sample volume, sample storage effects, and desorption/extraction efficiency.

The equivalent for the PFTIR is to observe three areas across the apparent width of the plume to ensure that the field of view of the instrument is filled by the plume. This also provides a measure of the gradient across the plume.

Determine known or suspected limitations (interferences and mixed matrix effects).

This was determined by analytical studies that preceded the field tests. Tests No. 4 and No. 5 of the plume generator test series were used to determine this through spectral residual measurements, if sufficient signal was observed for the interfering compounds.

During routine sample analysis, use control samples to prove the method and monitoring system performance are operating within acceptable limits.

Control sample equivalents are the flow metered gas mixture and the confirmatory in-stack EFTIR measurements in the plume generator tests.

Ensure each extraction batch QC consists, at a minimum, of samples to check bias, contamination and precision.

Potential bias was determined by Tests No. 1 through 5 of the plume generator test series. Contamination was evaluated using results of the heated air test periods between plume generator tests and from the triplicate heated air test conducted at the end of the plume generator test series. Precision was determined from the results of plume generator test Test No. 3 in which three replicate measurements were conducted.

For each batch, ensure a control duplicate is used to check instrument calibration and bias (calibration verification standard or laboratory control sample), blank(s) to assess contamination, and duplicates to measure precision and field samples.

The instrument is calibrated against a blackbody radiation calibration standard, which is NITS traceable. Radiation is calibrated against the Planck function for the controlled temperature of the source and it is known emissivity.

circumstances require that the horizontal path transmission be used to deduce gas concentrations in the ambient air, and the slant path transmission computed using typical atmospheric models and LBL calculations. This was done for all LBL calculations because it is an inherent part of the code. For CLS analyses, the measured horizontal-path transmission was used directly because the moderate height of the stack did not warrant separate computation of the slant path transmission.

The LBL inversion algorithm essentially iterates to generate an accurate temperature and transmission function for the plume during each spectral reduction. The CLS procedures used a manual process to estimate temperature from the ratio of adjacent H2O lines, which vary in

4-45

relative intensity with temperature. This manual temperature was used for all CLS analyses of spectra in the averaging interval for which temperature was estimated.

As was discussed in the procedures section, the measured background radiance needs to

be corrected for the plume transmission to provide the proper transmitted background signal, and the plume radiance must be fit to deduce the plume concentrations. In the LBL process, iterating on the transmission function allows the software to match the observed plume radiance corrected for both background and atmospheric path effects simultaneously. In the CLS method, the various measured spectra are combined to produce the plume radiance assuming a single isothermal isobaric plume path. The CLS algorithm then fits the measured absorbance to combinations of reference spectra to match the measured and computed absorbances. When the match is complete, the errors in the whole process can be estimated by comparing the spectral residuals between the measured and computed radiance spectra. The residuals in each of the analysis regions can then be converted to a maximum possible error on the gas concentrations. These errors were combined with those of the atmospheric path transmissions estimates to produce a maximum error on deduced gas concentrations.

References

1. D.M. Haaland, R.G. Easterling, Appl. Spectroc., 34, 539-548 (1980).

2. W.J. Phillips and G.M. Russwurm, “Open-Path FTIR Data Reduction Algorithm with Atmospheric Absorption Corrections: The Non-Lin Code,” in Proceedings of International Symposium on Industrial and Environmental Monitors and Biosensors, SPIE, Boston, MA., Nov. 1998

3. Bill Phillips, “IR Spectral Modeling and Reference Spectra Generation With E-Trans,” Air & Waste Management Association 94th Annual Meeting & Exhibition, Orlando, FL, June 2001.

5-1

5.0 Results

This section presents the results of the program to determine the accuracy and precision of the PFTIR method for measuring low concentrations of selected hydrocarbon species. These results were obtained from the plume generator tests and the controlled flare tests described in Section 4, and were used to calculate combustion efficiencies for the various test conditions. Section 5.1 summarizes the laboratory analytical study, which was conducted to calibrate and prepare the PFTIR for field testing. Sections 5.2 and 5.3 discuss the results of the plume generator and controlled flare tests, respectively. A description of the measurement uncertainty and quality control is presented in Section 5.4, and the statistical analysis of the results is provided in Section 5.5. 5.1 PFTIR Analytical Study The laboratory analytical study was undertaken, prior to field testing, to assess the performance expected from the PFTIR system in measuring hydrocarbons in flare exhausts. This study determined:

• Radiant signal levels expected for constituent gas species as a function of concentration and temperature, corresponding to various efficiency conditions and flare exhaust dilution;

• Minimum detection limits, and consequently maximum combustion efficiencies, that could be reliably calculated under various conditions; and

• Sensitivity of the signal to plume gradients.

Highlights of the laboratory analytical study are discussed below. Details of the study are

provided in Appendix C.

5.1.1 Radiant Emission Signal Levels Radiant emission signal levels were analytically simulated to provide estimates of the PFTIR signal levels that would be encountered in the field measurement program. To predict signal levels from a flare, radiant signature spectra for all significant chemical species were generated using a high temperature version of the HITRAN atlas in conjunction with rigorous software codes capable of constructing spectra from this atlas. Signal levels for all planned plume generator tests, defined in Section 4 of this report, were simulated.

5-2

Figure 5-1 shows the signature spectra expected from a high efficiency, low temperature flare, which is the difficult measurement case, and a low efficiency, high temperature flare, which is the easiest measurement case. As described in Section 4, the radiance is a function of concentration and temperature. A high efficiency flare will have low organic concentrations and this combined with a lower temperature will result in a weak signal that is more difficult to measure.

The spectra of Figure 5-1 indicate the signal levels that would need to be measured and

the radiances for each major compound are shown in Table 5-1. All signals are in the microwatt range and below. As seen in the table, the radiance for the low efficiency, high temperature case is approximately five times greater than that for the high efficiency, low temperature case.

Table 5-1. Radiance Levels Resulting from the Plume-Radiance Simulations Radiance (watts/cm2/str./cm-1)

Compound Low Efficiency High Temperature High Efficiency Low Temperature CO2 5.33 × 10-6 1.20 × 10-6 CO 0.90 × 10-6 0.194 × 10-6 THC 0.22 × 10-6 0.042 × 10-6

5.1.2 Minimum Detection Limits

To determine minimum detection limits for various species, laboratory simulations were conducted, as discussed above, for several chemicals at specific concentrations and at three temperatures: 150oC, 225oC, and 232oC.

Initially, the PFTIR system had to be calibrated to determine the radiance levels or equivalently the concentrations of the target compounds that the PFTIR instrument could detect. The calibration used a NIST traceable blackbody (as discussed in Section 4.1.4 Passive FTIR) because it has a known emissivity (0.95) and an accurately known temperature, it has a known radiance output. The calibration of the PFTIR provides a spectral response function for the instrument giving the radiance input per volt of system output. If the noise level of the PFTIR is measured, this noise level can be converted to a noise equivalent radiance or the noise limited minimum detectable gas concentration. Figure 5-1 shows an example of the radiance spectra corresponding to two sets of simulated flare conditions. The response function for the PFTIR, as measured in the laboratory, is shown in Figure 5-2.

5-3

Figure 5-1. Synthetic Radiance Spectra for a Low Efficiency, High Temperature Flare Simulation (top) and a High

Efficiency, Low Temperature Flare Simulation (bottom)

5-4

Figure 5-2. PFTIR System Response in (watts/cm2/str./cm-1) per Volt of FTIR Output. Note the 10-5 on the Y-axis

5-5

To determine the PFTIR noise level, the black body source was replaced by a liquid nitrogen cooled “cold plate.” The spectrum measured with this cold plate was considered the system noise level. The RMS noise levels seen in each of the three major analysis bands were: 1.33X10-8 at 1000 to 1025 cm-1, 1.70X10-8 at 2500 to 2600 cm-1 and 3.3X10-8 at 3300 to 3543 cm1. Table 5-2 shows the peak radiance levels determined from reference spectra for CO, ethylene, propylene, propane, and butane at specific concentrations and at three temperatures. These radiance levels translate into the noise-limited detections shown in the sixth and seventh columns of the table. Two spectral ranges are shown for propane because its signature is very weak in the 1000 cm-1 region (that is used for analysis of most other compounds), making it difficult to measure there. Since its band is much stronger in the 3000 cm-1 region, the detection of propane can be enhanced by using the shorter wavelength (larger wavenumber) band.

To further demonstrate the improvement that could be achieved using a different detector

(than was used for the field tests), the table also shows THC and THC-InSb for the 3000 cm-1 region. Here the situation is similar to that of propane except no reasonable signal is available at 1000 cm-1 for estimating total hydrocarbon. As a result it must be measured in the C-H stretch region around 3000 cm-1. The value shown for THC is what would be obtained using the MCT detector at 3000 cm-1. The value shown for THC-InSb is an estimate of the improvement in detection limit possible if an Indium Antimonide (InSb) detector were used. The InSb detector has a sensitivity one to two orders of magnitude better than the MCT so it provides better detection. However, the InSb is non-responsive at longer wavelengths, so it cannot be used alone. It would be necessary to use a “sandwich” detector, with the InSb on top and the MCT on the bottom, so that both would be used simultaneously each responding to different wavelengths.

The laboratory simulations showed that noise limited detections of less than 1 part per million per meter of pathlength (ppm-m) should be possible at 225°C for all compounds except propane. The 225°C (498K) condition corresponds to the “high temperature” cases in the plume generator tests. With the detection limits shown for this temperature the maximum detectable efficiency would be 99.96 percent. At 150°C the detection limits are higher so the maximum combustion efficiency detectable is lower at 99.95 percent.

5-6

Table 5-2. Minimum Detectable Gas Concentrations from Simulations at 150°C, 225°C, and 232°C

Peak Radiance

Noise limited (ppm*m)

900-1000 cm-1 3000 cm-1

Ref. Conc. (ppm)

Ref. Path (m) 1000 cm-1 3000 cm-1

150°C CO 2.50E-07 20 1 1.360 (C2H4) 7.03E-07 9.9 1 0.187 (C3H6) 1.39E-06 75 1 0.718 (C3H8) 6.58E-09 2.62E-08 20 1 40.426 25.19 (C4H10) 1.39E-06 73 1 0.698 THC 8.92E-08 30 2 20.20 THC - InSb* 2.22

225°C CO 3.50E-07 20 1 0.971 (C2H4) 1.17E-06 9.9 1 0.113 (C3H6) 2.30E-06 75 1 0.434 (C3H8) 1.00E-08 1.20E-07 20 1 26.600 5.50 (C4H10) 2.30E-06 73 1 0.422 THC 4.08E-07 30 2 4.86 THC - InSb* 0.49

232°C CO 3.80E-07 20 1 0.895 (C2H4) 1.20E-06 9.9 1 0.110 (C3H6) 2.37E-06 75 1 0.421 (C3H8) 1.06E-08 1.34E-07 20 1 25.094 4.93 (C4H10) 2.33E-06 73 1 0.417 THC 4.60E-07 30 2 4.31 THC - InSb* 4.60E-07 30 2 0.431 a Noise levels from LN2 source measurements (Watts/cm2/str/cm-1) * Estimated improvement in detection based upon InSb detector sensitivity compared to HgCdTe.

The results of the analytical study suggested that detection of propylene and total hydrocarbon would be substantially improved if an IbSb/HgCdTe sandwich detector were used rather than a standard HgCdTe detector. Unfortunately, the timing of the program forced the analytical study to start later than expected and only a month existed between the analytical work and the beginning of the field tests. Because sandwich detectors are special order items, one could not be obtained faster that about 90 days, making it impossible to incorporate it into the current program. The HgCdTe detector selected for use did have a peak response near 4.0 µm, which is somewhat shorter most detectors of this type. Its use was the best compromise available given the tight scheduling. Any follow on to this program, should use a sandwich detector to optimize the detection of propylene and other heavy VOCs.

5-7

5.1.3 Plume Gradients Temperature gradients occur from heat transfer losses and mixing of the original plume

contents with cooler ambient air as the plume rises. At the mouth of the plume generator stack where measurements were collected, a sharp gradient near the edge of the plume was expected with a small mixing region. For the controlled flare, turbulent mixing due to higher stack discharge velocity is expected to result in a more pronounced gradient near the edge of the plume, but might result in a lesser gradient within the plume. To examine the effect of extreme gradients on the radiant signature, the three temperature profiles shown in Figure 5-3 were modeled. Profile 1 is an example of very extreme cooling of the plume and not expected to occur in either the plume generator or in an actual flare. Profile 2 is more moderate and could occur in the simulator, although a penetration of nearly 25 percent into the plume is more than expected. Profile 3 is considered the most likely to occur in the simulator.

Radiant signatures produced by the three profiles are shown in Figure 5-4. The most

pronounced effect is a lowering of the effective temperature along the line of sight, as seen for Profiles 1 or 2, which reduces the radiant emissions. But the emissions seem to be reduced proportionally throughout the spectrum; therefore, the ratio of the various gas concentrations may not be strongly affected. A profile can be taken into account with the full line-by-line (LBL) calculations. This should result in a correct combustion efficiency even with severe gradients. If the gradients are found to be significant to actual flare data, measurement of these gradients may become important to the determination of the combustion efficiency.

As a test of the significance of the gradient on the estimated combustion efficiency,

analyses were used to analyze spectra from each of the gradients. The algorithm assumed a single isothermal zone and estimated CO, CO2, and THC. The results of the analyses are shown in Table 5-3. The temperatures derived from each simulated spectrum are shown as the “effective temperature” (Eff. Temp.). These are the rotational-temperatures derived from plots of the CO spectral line intensities seen in the spectra versus rotational quantum number for CO. An example of this is shown in Figure 5-5 for the severe roll of Profile 1. As seen in the table, the effective temperatures deduced for Profiles 2 and 3 were close to the actual temperature, indicating that the mild gradients did not alter the observed spectra severely. However, the severe roll-off of Profile 1 did produce errors in temperature estimates because of alterations to the observed spectra.

5-8

Temperature Radial DistributionTest Case 1

300

350

400

450

500

550

0.00 10.00 20.00 30.00 40.00 50.00Radial Position, cm

Tem

pera

ture

, K

Profile1Profile 2Profile 3

Figure 5-3. Three plume Temperature Gradients Modeled During the

Figure 5-4. Calculated CO2 and CO Radiance Levels at 498K for the Three Profiles

of Figure 5-3

5-9

Table 5-3. Analysis of Spectra Generated Using the Three Profiles Shown in Figure 5-3

Profile 1 Profile 2 Profile 3 Conc. (ppm) Error Conc. (ppm) Error Conc. (ppm) Error CO 26.0 0.7 41.1 0.865 53.8 1.00 CO2 9166 540 10281 577.13 11085 598.7 HC 12.1 0.2 20.9 0.374 28.2 0.497 Effective Temp. 214 225 227 Efficiency (%) 99.59 99.40 99.27

As seen in Table 5-3, the derived combustion efficiencies did not vary significantly compared to the observed concentration variations. This implies that the reduction in radiance was very nearly equal in the CO, CO2, and THC spectral regions so the influence of the gradient is not strong for deduced efficiency even though it is for individual gas concentrations. The field test results on the plume generator and the controlled flare confirmed that the effect of the gradient was insignificant. 5.2 Plume Generator Testing

Plume generator tests, the bases of which were discussed in Section 4 were conducted over a 3-day period in August 25-27, 2003, as shown in Table 5-4. As seen in the table, there are two sets of test data, the “a” set and the “b” set. The target gas spiking rates were the same for both sets of tests. The difference in the two is the manner in which the PFTIR data were collected. In the “a” series of tests, all the PFTIR data were collected at the centerline of the plume. In the “b” tests, the PFTIR was traversed across the plume so that three or more “slices” of the plume were taken. The purpose of taking these slices was to provide data for the temperature and concentration gradients at the edge of the plume. The results of these tests showed that the gradients were minimal, so that the centerline data could be used exclusively for the analyses.

5-10

y = -5.62694E-03x - 3.32708E+01R2 = 9.94491E-01

-35.5

-35

-34.5

-34

-33.5

-330 50 100 150 200 250 300 350 400

J'(J'+1)

Log(

L/(n

u^4*

m))

Figure 5-5. Log plot of CO rotational-line energy versus line quantum number.

The slope of this plot gives the effective gas temperature, which in this case is 213.9 °C.

5-11

Table 5-4. Test Schedule for the Plume Generator Tests

Test No. Description Day Date Testing Interval

Set-up for tests Monday 8/25 - Ambient air Monday 8/25 14:20 – 14:49 Heated ambient air Monday 8/25 15:00 – 15:58 Stratification test with CO spiking Monday 8/25 15:59 – 16:28

1a Low Eff. /High Temp., butane only, centerline PFTIR

Tuesday 8/26 9:56 – 10:09

2a Mid Eff. / High Temp., mixed gases, centerline PFTIR

Tuesday 8/26 10:23 – 10:38

3a High Eff. / High Temp., mixed gases, centerline PFTIR

Tuesday 8/26 10:49 – 11:02

4a Mid Eff. / Low Temp., mixed gases, centerline PFTIR

Tuesday 8/26 13:08 – 13:21

5a High Eff. / Low Temp., mixed gases, centerline PFTIR

Tuesday 8/26 13:37 – 13:51

4b Mid Eff. /Low Temp., butane only, multi point PFTIR

Tuesday 8/26 14:04 – 14:16

5b High Eff. / Low Temp., mixed gases, multi point PFTIR

Tuesday 8/26 14:40 – 14:56

2b Mid Eff. / High Temp., mixed gases, multi point PFTIR

Tuesday 8/26 15:42 - 15:56

3b-1,2,3 High Eff. / High Temp., mixed gases, multi point PFTIR, triplicate tests

Tuesday 8/26 16:13 – 17:10

1b Low Eff. / High Temp., mixed gases, multi point PFTIR

Wednesday 8/27 10:28 – 11:07

Heated ambient air, multi point PFTIR, triplicate tests

Wednesday 8/27 11:13 – 11:46

5.2.1 Stratification Tests

For the plume generator tests, the gas species were injected into a duct containing the heated air. Due to the length and configuration of the ducting arrangement, it was believed that the spiked gases would be thoroughly mixed with the hot air prior to the sampling location in the stack. It was necessary to have complete mixing so that the EFTIR data could be collected at a single point, and be representative of the concentration that the PFTIR would measure at the outlet of the stack. Therefore, a stratification test was conducted in the plane of the sampling location to confirm that sufficient mixing had occurred.

The CO concentration was measured by the EFTIR, and was recorded at four different

locations within a 36-inch diameter stack. The testing positions were located along a single axis at points 24 inches, 18 inches, 12 inches, and 6 inches from the inner wall of the 36" diameter

5-12

stack. Two tests were performed at the 18-inch location. A total of five tests were performed during consecutive time intervals. The target CO concentration for each of the tests was fifteen parts per million on a volume basis (ppmv). The concentrations of ambient H2O and CO2 were also measured. The ambient H2O and CO2 concentration measurements were included because they are completely mixed, and so variations in those measurements could be attributed to the measurement, and not to the degree of mixing.

The data from the stratification testing are presented in Table 5-5. The data in the table are presented in chronological order, and are separated by the location in the stack at which they were collected. The measurements of CO2 and H2O are given as percent by volume, while the data for CO are provided in ppmv. In addition, the average and standard deviation of the CO data for each test location is included. The mean values of all the measurements of CO, CO2, and H2O are also presented, along with the standard deviation of all the values for CO. It should be noted that all of the CO data fall with twice the standard deviation, and that the 2σ value represents the 95 percent confidence interval for the CO measurement.

The CO, CO2, and H2O data were all normalized to their respective average values for all of the tests, so that it would be possible to compare the variability of the test data at each location in the stack. These results are presented in Figure 5-6. The x-axis of the figure denotes the location in the stack, in inches from the sample port along the wall of the stack. The radius of the stack is 18 inches, and the centerline of the stack is labeled in the figure. The two sets of data that were collected at the 18-inch point were offset slightly from the centerline so that the differences could be more easily discerned.

In the figure, the data for the normalized CO2 measurements is represented by triangles, while the normalized H2O results are represented by circles. The CO data are presented as solid diamonds. The data clearly show that the scatter of the CO is not any greater than that of the CO2 or H2O. Since the ambient concentrations of these gases are mixed completely, one can therefore conclude that the CO that was injected into the hot air stream is also mixed completely.

5-13

Table 5-5. EFTIR Stratification Test Data

H2O CO2 CO Average for CO

Standard Deviation

for CO Date & Time % % ppmv

Begin at Point #2 18"in Stack 4:08:29 PM 2.118 0.0393 15.03 4:09:11 PM 2.159 0.0397 15.23 4:09:52 PM 2.143 0.0416 15.41 4:10:34 PM 2.137 0.0404 15.47 4:11:16 PM 2.144 0.0409 15.60 4:11:58 PM 2.189 0.0409 15.75 4:12:40 PM 2.214 0.0408 15.60 15.44 0.245

Moved to Point #3 12" in Stack 4:13:22 PM 2.184 0.0399 15.37 4:14:04 PM 2.131 0.0408 15.37 4:14:46 PM 2.139 0.0403 15.30 4:15:28 PM 2.151 0.0396 15.11 4:16:10 PM 2.089 0.0404 15.18 4:16:52 PM 2.139 0.0390 15.10 15.24 0.124

Moved to point #4 6" in Stack 4:17:34 PM 2.094 0.0395 14.96 4:18:16 PM 2.115 0.0386 15.07 4:18:58 PM 2.005 0.0379 14.71 4:19:40 PM 1.974 0.0362 14.62 4:20:21 PM 1.957 0.0420 14.68 4:21:03 PM 2.174 0.0380 15.36 14.90 0.285

Moved to point #1 24" in Stack 4:21:45 PM 2.181 0.0384 15.38 4:22:27 PM 2.025 0.0367 14.62 4:23:09 PM 2.016 0.0364 14.68 4:23:51 PM 2.049 0.0379 14.88 4:24:33 PM 2.228 0.0396 15.18 4:25:15 PM 2.273 0.0393 15.32 15.01 0.329

Repeat Point #2 18"in Stack 4:25:57 PM 2.222 0.0409 15.37 4:26:39 PM 2.293 0.0403 15.72 4:27:21 PM 2.324 0.0404 15.72 4:28:03 PM 2.219 0.0411 15.61 4:28:45 PM 2.210 0.0403 15.48 15.58 0.153 Mean for all tests 2.143 0.0396 15.23 0.340

5-14

Figure 5-6. Stratification Tests Results

0.90

0.93

0.95

0.98

1.00

1.03

1.05

1.08

1.10

0 6 12 18 24 30 36

Stack Location, inches

Nor

mal

ized

Con

cent

ratio

n

CO2 H2O

CO

Stack Centerline

5-15

5.2.2 Emissions Tests Measurements of the emissions from the plume generator were performed according to the test program that was developed as part of the QAPP. The following sections summarize the results of those measurements, while the detailed data are presented in the appendices. 5.2.2.1 Concentration Measurement Data Plume generator tests were conducted to determine the ability of PFTIR to measure known concentrations of spiking gases in a heated air stream discharged from the plume generator stack. Mixtures were chosen to simulate the composition of actual flares. The PFTIR measured the gases leaving the stack. The “known” concentrations were determined from in-stack measurements by three methods:

• Process data and calculated values of spiking gas concentrations;

• Canister sample measurements for one of the five test series (Test No. 3a, in Table 5-4); and

• EFTIR measurements of in-stack gas composition.

The intention of the test program was to use both the process gas flow data and the

EFTIR measurements as continuous reference methods to which the PFTIR measurements could be compared. The canister measurements were performed as a single-point validation of the accuracy of the continuous reference method data. However, the process flow data were found to produce spiked gas concentrations with a bias, due to leakage of the hot air out of the system downstream of the air flow measurement location, and upstream of the point where the spiked gases were injected. In addition, due to a miscommunication, the process flow data were not collected continuously during the testing. Therefore, for the plume generator tests, the EFTIR measurements are designated as the single reference method to which the PFTIR is being compared.

Table 5-6 compares process data-calculated and EFTIR-measured plume generator gas

compositions, by test. These data show that for most tests, the values determined by the EFTIR are about five percent higher than those calculated from process flow measurements. The exceptions to this trend are the CO and propane values for Test 1b. It should also be noted that the ambient CO2 concentration was about 400 ppmv. As a result, the CO2 values measured by the EFTIR would be expected to be approximately 400 ppm higher than that calculated for using the process flow data for the spiked gases.

5-16

Table 5-6. Plume Generator Tests - Spiking Gas Concentrations by Calculation and EFTIR Measurement Process Data - Calculated Plume Generator

Stack Concentration (ppmv)

Average EFTIR Measured Concentrations a

(ppmv)

Test Condition Date

Process Data

Sample Time C

O

CO

2

But

ane

Eth

ylen

e

Prop

ylen

e

Prop

ane

CO

CO

2

But

ane

Eth

ylen

e

Prop

ylen

e

Prop

ane

Stratification 8/25/2003 3:59:44 13 0 0 0 0 0 13 c 400 c BDL BDL BDL - Startup 8/26/2003 9:45:18 0 6,070 0 0 0 0 NA NA NA NA NA - Test 1a 8/26/2003 9:56:39 141 6,005 108 0 0 0 144 7270 83 BDL BDL - Test 2a 8/26/2003 10:25:50 71 6,011 54 27 28 0 74 6600 55 29 36 - Test 3a 8/26/2003 10:49:25 21 6,002 22 16 17 0 24 6750 22 18 21 - Temp. change 8/26/2003 12:47:22 0 5,994 0 0 0 0 NA NA NA NA NA - Test 4a 8/26/2003 1:09:23 72 6,015 55 27 28 0 74 6970 64 29 34 - Test 5a 8/26/2003 1:39:13 22 6,031 21 16 16 0 24 6980 26 18 21 - Test 4b 8/26/2003 2:15:11 71 5,988 0 30 28 0 74 6580 63 28 33 - Test 5b 8/26/2003 2:54:04 22 6,020 22 17 16 0 24 6740 27 18 20 - Test 2b 8/26/2003 3:45:13 72 6,092 55 28 28 0 73 6960 63 28 34 - Test 3b-1 8/26/2003 4:09:37 23 6,076 22 17 17 0 24 6340 27 18 20 - Test 1b 8/27/2003 10:46:52 99 5,933 0 0 0 80 b 138 6570 NA BDL BDL 105 Heated air tests 8/27/2003 11:33:17 0 0 0 0 0 0 NA NA NA NA NA - a Values represent average concentrations (unless noted) over the time intervals used to reduce the PFTIR data for each test. The Process

Data Sample Time represents the time at which the instantaneous spiking gas flow measurement data were collected for each test. b Butane supplies were exhausted; therefore, commercial grade propane from the Zink fuel supply system was substituted for butane during

this test. c Reported values represent instantaneous EFTIR values rather than an average. Average EFTIR values for the stratification tests were not

calculated. BDL = Below Detection Limit. NA = Not Applicable or Not Available depending on condition.

5-17

During the designated startup and heated air tests, no spiking gases were fed to the system. Only CO was fed during the stratification test (the value of CO2 measured by the EFTIR is the ambient concentration). Measurements for these parts of the tests are discussed further in Section 5.2.1 of this report.

For comparison, Table 5-7 presents the spiking gas concentrations for EFTIR and PFTIR using LBL data reduction. The EFTIR method is a standard EPA method, and the results are used in this study as a measure of the “true” composition of the plume generator gas stream used for the PFTIR evaluation. The percent error of the PFTIR measurement for each data pair is also presented in the table. The average values of the concentration measurements are shown for all spiking tests, designated by the test numbers. These correspond to the test conditions defined in Section 4. The table also shows the date and the start time of each test.

The test plan was developed in an effort to provide insight into the effect of temperature

and concentration on the accuracy of the PFTIR. Since signal strength is proportional to both of these parameters, it was expected that the accuracy of the high temperature test conditions would be better than that at the low temperature if the system were signal-to-noise limited. Similarly the results at the higher concentration conditions (simulating lower destruction efficiency) should be better than those at the low concentration (high destruction efficiency). Using that logic, test condition 2 would be expected to have the best agreement between the PFTIR and EFTIR, while test condition 5 should have the worst.

Inspection of the data in Table 5-7 illustrates how these parameters had a varied impact,

depending upon the gas species. For CO, the PFTIR measurements agreed with the EFTIR measurements better at the higher temperature. Ironically, these measurements agree better with each other than with the canister samples that were collected during Test 3b (which was a high temperature test). Contrary to what was expected, the results were not better at higher concentrations of CO. The data for the remaining gases did not demonstrate any discernable trends. Thus, these data did not validate the hypothesis that the measurements should agree better for Test 2 than for Test 5.

5-18

Table 5-7. Plume Generator Tests - Comparison of Average Gas Concentrations

CO CO2 Butane Ethylene Propylene Propane

TEST E

FTIR

(ppm

v)

PFT

IR (L

BL

) (p

pmv)

% E

rror

EFT

IR (p

pmv)

PFT

IR (L

BL

) (p

pmv)

% E

rror

EFT

IR (p

pmv)

PFT

IR (L

BL

) (p

pmv)

% E

rror

EFT

IR (p

pmv)

PFT

IR (L

BL

) (p

pmv)

% E

rror

EFT

IR (p

pmv)

PFT

IR (L

BL

) (p

pmv)

% E

rror

EFT

IR (p

pmv)

PFT

IR (L

BL

) (p

pmv)

% E

rror

Test 1a 144 113 22 7,270 3,549 51 83 23 72 BDL -1.5 - BDL 44 - - - - Test 2a 74 65 12 6,600 4,017 39 55 4 93 29 18 38 36 60 67 - - - Test 3a 24 23 4 6,750 3,365 50 22 -1 18 10 44 21 53 152 - - - Test 4a 74 112 51 6,970 8,123 17 64 16 75 29 27 7 34 96 182 - - - Test 5a 24 46 92 6,980 7,996 15 26 25 4 18 15 17 21 85 305 - - - Test 4b 74 96 30 6,580 6,368 3 63 17 73 28 22 21 33 85 158 - - - Test 5b 24 45 88 6,740 8,238 22 27 26 4 18 15 17 20 79 295 - - - Test 2b 73 91 25 6,960 6,354 9 63 20 68 28 22 21 34 62 82 - - - Test 3b-1 24 29 21 6,340 5,700 10 27 7 74 18 11 39 20 43 115 - - - Test 3b-2 24 25 4 6,340 4,575 28 27 4 85 18 10 44 20 46 130 - - - Test 3b-3 24 26 8 6,330 4,663 26 29 5 83 18 11 39 11 a 40 264 - - - Test 1b 138 - - 6,570 - - - - - BDL - - BDL - - 105 b - -

a CO supplies were exhausted near the end of this test period. b Butane supplies were exhausted, therefore commercial grade propane from the Zink fuel supply system was substituted for butane

during this test. BDL = Below Detection Limit.

5-19

It should be noted that the negative concentrations that were determined by the PFTIR are an artifact of the analytical method, in which a variable baseline was fit to account for possible calibration variations. Negative values could have resulted from this fitting parameter but they can also occur if interferant features in the spectrum have the appearance of the inverse of the analyzed compound. This can occur, for example, if strongly absorbing species are in the same spectral region as the one that is used for detection of an analyte, and these interferants are not accurately accounted for with the references used. In a typical application, these values would be reported as either zero, or BDL, but they were preserved in the table to illustrate the phenomenon.

The data in Table 5-7 clearly demonstrate that the results of the PFTIR measurements are

repeatable. This is most evident for Tests 3b-1, 3b-2, and 3b-3, which were performed as part of the precision test, described in Section 5.2.3. Comparisons of Tests 4a and 4b, as well as Tests 5a and 5b, also indicate good repeatability of the PFTIR data.

Since the PFTIR measurements for CO appear to be reasonably accurate, this would

suggest that the method provides a promising means for determining combustion efficiency, in which the CO concentration is the dominant term. However, the values for individual gas species are less accurate, which would have an adverse effect on estimates of speciated mass emissions or destruction efficiency. One cause for optimism is that the plume generator was operating at a gas temperature lower than what would be found in a typical operating flare. The higher temperature will produce a stronger signal and improve the accuracy of the analysis. There are also higher resolution detectors available, which could improve the strength of the signal by orders of magnitude.

Canister samples, taken in triplicate for Test 3a, were used as backup validation samples

for the EFTIR. These data are presented in Table 5-8. As described in Section 4.1.3.1, the samples were collected immediately upstream of the inlet of the EFTIR sample cell. The data show that the canister measurements agree well with those obtained using the EFTIR.

5-20

Table 5-8. Plume Generator Test 3a EFTIR Canister PFTIR CO 24 34 23 CO2 6750 6600 3365 Butane 22 25 -1 Ethylene 18 22 10 Propylene 21 25 53

The PFTIR data are more mixed in their agreement with the EFTIR results. For example,

the CO values are very accurate, while the CO2, ethylene, and propylene results show significant variation. The CO, CO2, and ethylene all meet the target accuracy for individual species of ±50 percent. However, as described previously, the detector that was used for the tests did not have a strong signal in the part of the spectrum in which the butane signal occurs. As a result, the PFTIR did not detect butane very well (a feature that is correctible with a different detector). Again, the negative value obtained by the PFTIR for butane is an artifact of the analysis and is analogous to a reading that is below the detection limit of the instrument.

Since the canister data provides additional reference data for comparison to the PFTIR measurements, Figures 5-7 through 5-11 have been included to illustrate the data from the table. The time series data for each gas species has been provided to highlight the comparison of the EFTIR and canister results, with those from the PFTIR. The figures demonstrate that the best agreement among the methods occurs in the CO data, while the other species show less agreement. An interesting feature of the canister results is that they agree quite well with the EFTIR results, except in the case of the CO data. The source of this discrepancy is unknown.

Another feature that is apparent from Figures 5-7 through 5-11 is that the PFTIR data has

more variability than that for the EFTIR. This would be expected, since the EFTIR measurements were made in the stack, while the PFTIR measurements were made above the stack, and subject to factors such as wind speed.

5-21

Figure 5-7. Time Series for CO2

Test 3a: Carbon Dioxide

2000

3000

4000

5000

6000

7000

8000

10:48:00 10:50:53 10:53:46 10:56:38 10:59:31 11:02:24 11:05:17Time on 8/26/03

Con

cent

ratio

n (p

pmv)

PFTIR-LBL

Canister

EFTIR

5-22

Figure 5-8. Time Series for CO

Test 3a: Carbon Monoxide

0

10

20

30

40

50

60

10:48:00 10:50:53 10:53:46 10:56:38 10:59:31 11:02:24 11:05:17Time on 8/26/03

Con

cent

ratio

n (p

pmv)

PFTIR-LBL

Canister

EFTIR

5-23

Figure 5-9. Time Series for Butane

Test 3a: Butane

-10

0

10

20

30

40

50

60

10:48:00 10:50:53 10:53:46 10:56:38 10:59:31 11:02:24 11:05:17Time on 8/26/03

Con

cent

ratio

n (p

pmv)

PFTIR-LBL

Canister

EFTIR

5-24

Figure 5-10. Time Series for Ethylene

Test 3a: Ethylene

0

10

20

30

40

50

60

10:48:00 10:50:53 10:53:46 10:56:38 10:59:31 11:02:24 11:05:17Time on 8/26/03

Con

cent

ratio

n (p

pmv)

PFTIR

Canister

EFTIR

5-25

Figure 5-11. Time Series for Propylene

Test 3a: Propylene

0

10

20

30

40

50

60

70

80

90

100

10:48:00 10:50:53 10:53:46 10:56:38 10:59:31 11:02:24 11:05:17Time on 8/26/03

Con

cent

ratio

n (p

pmv)

PFTIR

Canister

EFTIR

5-26

See Section 5.4 for a discussion of measurement uncertainty and quality control, and Section 5.5 for the results of the statistical analyses of the test data. More details on the plume generator test data, including time-series data sets, are presented in Appendix B (EFTIR data) and Appendix C (PFTIR data). 5.2.2.2 Combustion Efficiency Results

Table 5-9 compares combustion efficiencies determined from the PFTIR and EFTIR data as well as target values for the plume generator tests. The target combustion efficiency represents the calculated efficiency based on the target concentrations for the various compounds as defined in the project QAPP. The EFTIR combustion efficiency is assumed to represent the “actual” simulated efficiency for each test. The uncertainty terms included with the PFTIR data were determined from the statistical analysis, and are presented in more detail in Section 5.5. The specific values that were used for these data are presented in Table 5-20. The EFTIR and PFTIR data are also presented graphically in Figure 5-12.

The combustion efficiency is defined in the QAPP as the concentration of CO2 divided by

the sum of the concentrations of the CO2, CO, THC, and soot. For a gas-fired source applying good combustion practice, the soot emissions should be negligible (none was observed during any of the tests at John Zink). As described previously, the PFTIR was not able to determine a THC value from the H-C stretch of the spectrum. As a result, for these combustion efficiency calculations, the sum of the concentrations of the individual hydrocarbon species was used in place of a THC value.

As seen in Table 5-9, the uncertainty of the EFTIR measurement was significantly lower

than that for the PFTIR, as expected. The graphical presentation in Figure 5-12 demonstrates that the error bands for the PFTIR data overlapped those of the EFTIR for about half of the tests.

The data in the table suggest that the combustion efficiency determined by the PFTIR

agreed well with the EFTIR values. However, this may simply be a fortunate circumstance. The combustion efficiency calculation is dominated by the terms for CO2 and CO, which the PFTIR was found to predict with acceptable accuracy. The data in Table 5-7 revealed that the hydrocarbon species were not predicted as well. These results showed that the ethylene values generally had a negative bias, while the propylene had a significant positive bias, and the butane was not detected well at all. By summing the hydrocarbon concentrations, these biases cancelled each other out to produce what appears to be an accurate value for combustion efficiency.

5-27

Table 5-9. Simulated Combustion Efficiency Comparison Between EFTIR and PFTIR Measurement Calculations

Test Process Condition Target

Efficiency

Average EFTIR

Efficiency Average PFTIR-LBL Efficiency

1a Low Efficiency / High Temp. 96.0 97.0 ± 0.12 95.2 ± 0.30 1b Low Efficiency / High Temp.

(propane substituted for butane) 96.0 96.4 ± NA NA a

2a Mid Efficiency / High Temp. 97.1 97.1 ± 0.04 96.4 ± 0.35 2b Mid Efficiency / High Temp. 97.1 97.2 ± 0.03 97.0 ± 0.12 3a High Efficiency / High Temp. 98.7 98.8 ± 0.04 97.5 ± 0.29

3b-1 High Efficiency / High Temp. 98.7 98.6 ± 0.02 98.4 ± 0.11 3b-2 High Efficiency / High Temp. 98.7 98.6 ± 0.03 98.2 ± 0.12 3b-3 High Efficiency / High Temp. 98.7 98.7 ± 0.08 98.3 ± 0.06 4a Mid Efficiency / Low Temp. 97.1 97.2 ± 0.04 97.0 ± 0.39 4b Mid Efficiency / Low Temp. 97.1 97.1 ± 0.11 96.5 ± 0.82 5a High Efficiency / Low Temp. 98.7 98.8 ± 0.03 97.9 ± 0.25 5b High Efficiency / Low Temp. 98.7 98.7 ± 0.03 98.0 ± 0.53

a NA = Not Available. Efficiency could not be calculated from the PFTIR data because the signal for propane was too low and CO spiking gas supplies were depleted shortly after this test began.

Based on the detection limit data that were presented in Section 5.1.2, the PFTIR data can

be used to calculate combustion efficiencies in the ranges needed for flare evaluations. However, the detector that was used for the plume generator tests did not provide sufficient signal strength to produce a value for THC. In addition, the analyses of the concentration for the other hydrocarbon species were hampered by low signal strength. As described previously, a new detector, or combination of detectors, may be implemented as part of future tests to optimize the method and provide more accurate values for the concentration of both THC and the individual hydrocarbon species. It is believed that with both hardware and analysis method optimization, the PFTIR results can be substantially improved.

5-28

94.0

95.0

96.0

97.0

98.0

99.0

100.0

1a -

EFTI

R

1a -

PFTI

R

2a -

EFTI

R

2a -

PFTI

R

2b -

EFTI

R

2b -

PFTI

R

3a -

EFTI

R

3a -

PFTI

R

3b -1

- EF

TIR

3b -1

- PF

TIR

3b -2

- EF

TIR

3b -2

- PF

TIR

3b -3

- EF

TIR

3b -3

- PF

TIR

4a -

EFTI

R

4a -

PFTI

R

4b -

EFTI

R

4b -

PFTI

R

5a -

EFTI

R

5a -

PFTI

R

5b -

EFTI

R

5b -

PFTI

R

Test Sequence for EFTIR and PFTIR

CE

(%)

Measured Average Efficiency%Analytical Error Upper Bound %Analytical Error Lower Bound %

Figure 5-12. Calculated Combustion Efficiencies with Uncertainty Bars for EFTIR and PFTIR

5-29

5.2.3 Precision Test A special precision test was run at Test 3 conditions on the plume generator. The

purpose of the precision test was to demonstrate the repeatability of the PFTIR measurements over time. The triplicate tests are designated as Tests 3b-1, 3b-2, and 3b-3. The results of these tests are presented in Table 5-10.

The data in the table show that the measured concentrations were consistent throughout

the testing period, with the exception of the results for CO2 and butane in Test 3b-1. These values are significantly higher than their respective averages. As a result, the coefficient of variance is much higher for the CO2 and butane than for the other compounds that were measured. The combustion efficiency data suggest that the repeatability of this calculation is quite good. But, the CO2 will dominate the efficiency calculation. The increased efficiency in test 3b-1 is therefore driven by the increased CO2 observed and not by the THC.

Table 5-10. PFTIR Results for Test 3b Replicate Measurements

Measured Concentrations (ppmv) Test CO CO2 Butane Ethylene Propylene

Combustion Efficiency (%)

3b-1 29 5,700 7 11 43 98.4 3b-2 25 4,575 4 10 46 98.2 3b-3 26 4,663 5 11 40 98.3

Average 27 4,980 5.3 11 43 98.3 Std Dev. 2.1 626 1.5 0.6 3 0.08 CV (%) 7.7 13 29 5.4 7.0 0.1

CV = Coefficient of Variation

5.3 Controlled Flare Testing The controlled flare PFTIR test was conducted over a 45-minute test period on August

27. The purpose of this test was to assess the logistical challenges associated with performing the PFTIR measurements on an actual flare. These challenges included the presence of a reaction zone (in which the gas compositions would be changing), and the dynamic motion of an open flame exposed to ambient conditions.

A chronological account of the test activities is presented in Table 5-11. As seen in the

right-hand column of the table, the location of the PFTIR scans was varied significantly. The horizontal views extended from one extreme edge of the plume to the other. The vertical views

5-30

extended from a location well above the combustion zone down to a point in which there were flamelets entering the field of view. Thermal imagery was used to locate the PFTIR field of view relative to the flare gases.

Two canister grab samples of the commercial-grade propane used to fire the flare were

collected at intervals during the flare test and analyzed to characterize the fuel composition. A discussion of the results follows.

Table 5-11. Measurement Activities for the Controlled Flare Test (August 27)

Time Activity PFTIR Measurement Point 17:01 Flare setup and pretests 17:12 Start flare

17:13:17 Start PFTIR Acquisition 17:25:33 Vertical view Just above combustion zone 17:28:36 Vertical view Higher above combustion zone 17:33:33 Pause PFTIR adjust tripod

17:30 Fuel canister sample collected 17:36:36 Restart PFTIR 2 m above combustion zone 17:39:44 Horizontal views Start at left extreme of plume and move right17:41:55 Horizontal views Centerline of plume 17:43:54 Horizontal views Right of plume centerline 17:47:16 Horizontal views Back to plume centerline 17:53:18 Vertical views Lower until flamelets getting into FOV

17:50 Fuel canister sample collected 17:56:18 Vertical views Lower just above flame zone, centerline 17:57:39 Horizontal views Left of center just above flame zone 17:59:37 Horizontal views Right of center just above flame zone 18:02:11 Flare shutdown 18:03:01 Sky background 18:06:08 Stop sky background

5.3.1 Flare Fuel Analysis

Sample results for the propane fuel samples are summarized in Table 5-12, and compared to a fuel supplier specification that was provided to the John Zink test facility on August 7, 2003.

Results for Sample 1 were considered invalid because of discrepancies between the final

canister vacuum reading obtained during sample collection (-10 in. Hg) and the vacuum reading obtained by the lab upon receipt of the sample (-28 in. Hg). The most likely cause of this

5-31

discrepancy is a faulty pressure gauge reading during sample collection or in the lab, resulting in improper pressurization and dilution of the sample prior to analysis.

Table 5-12. Canister Sample Results for Propane Fuel

Sample 1 a (vol %)

Sample 2 (vol %)

Fuel Supplier Specification,

(vol %) Parameter 8/27/03, 17:30 8/27/03, 17:50 8/7/03

Methane 0.0068 0.04 NR Ethane 0.68 4.8 6.3 Propane 11 92 93.5 Isobutane 0.056 0.38 0.183 Butane ND (0.03) 0.022 0.019 Neopentane ND (0.03) ND (0.0025) NR Isopentane ND (0.03) ND (0.0025) NR Pentane ND (0.03) ND (0.0025) NR C6+ Hydrocarbons ND (0.3) ND (0.025) NR

a Sample results considered invalid. ND = Not detected (reporting limit shown in parentheses). NR = Not reported.

Results for Sample 2 were comparable to the fuel supplier analysis, indicating

approximately 92 vol% propane, 5 vol% ethane, 0.4 vol% butane plus isobutane, and 0.04 vol% methane (400 ppmv). Isomers of pentane and total C6+ hydrocarbons were not present at concentrations above the reporting limits of 25 ppmv and 250 ppmv, respectively.

5.3.2 Flare Passive FTIR Measurements

The controlled flare test was conducted to determine how to apply PFTIR to an actual flare and evaluate measurement performance for the current state of method development. The flare was fueled with commercial grade propane. The combustion air mixed with the fuel by induced entrainment at the flare stack tip.

The PFTIR equipment was located at a convenient distance from the flare and oriented to provide a line of sight to the region above the visible flame zone for the measurements. The distance above the visible flame zone varied from near zero to over 2 meters.

The PFTIR telescope encompassed a circular field of view in the region of interest 12 inches in diameter. Both centerline and traverse measurements were made at various times

5-32

during the overall test period. Data for plume efficiency estimates is based on centerline measurements made during four distinct test periods of about 1 to 3 minutes each.

The radiance spectral data were captured at approximately 1.5-second intervals and averaged over 30 seconds to represent one “sample data point.” A time series of 30-second averages over four separate test periods was used to generate the controlled flare test data. Table 5-13 summarizes the controlled flare-PFTIR data.

Table 5-13. Controlled Flare Test PFTIR Data Summary

Average Species Concentrations (ppmv)

Data Averaging

Period

Average Plume Temp (°C)

CO

CO

2

But

ane

Eth

ylen

e

Prop

ylen

e

Prop

ane Average

Combustion Efficiency

(EFF) %

Incomplete Combustion

Emission Factor

(1-EFF/100) 17:25:33 – 17:28:19

302 57 98,100 0.7 1.1 27 45 99.9 ± 0.30 0.0013 ± 0.30

17:28:36 – 17:31:07

293 71 132,300 1.4 0.45 29 140 99.8 ± 0.30 0.0018 ± 0.30

17:41:55 – 17:43:03

225 58 60,000 3.3 1.8 33 40 99.8 ± 0.30 0.0023 ± 0.30

17:55:46 – 17:57:39

416 390 248,200 2.5 1.9 43 1255 99.5 ± 0.30 0.0052 ± 0.30

The table presents data for the concentration of a number of compounds within the flare plume. The uncertainties that are included with the combustion efficiency values are the average uncertainty associated with the PFTIR measurements that were conducted on the plume generator. As seen in the figure, the uncertainty calculated for the controlled flare by the PFTIR is actually higher in most cases than the percent of incomplete combustion. Nonetheless, the data suggest that the combustion efficiency of the controlled flare is very high.

The different data averaging periods reflect different locations in the plume at which the

PFTIR was focused (see Table 5-11). For example, in the fourth row of data, the PFTIR was sighted at a point in the plume that was very close to the active flame zone. In fact, flamelets were seen to peel off of the main flame and enter the field of view during these tests. The higher temperature and higher gas concentrations both verify that this location was closest to the flame, and therefore the hot gases had not been cooled or diluted as much as in the other cases.

5-33

At this location, the combustion was incomplete and one would expect unburned gas concentrations to be higher. The data demonstrate this clearly for the CO and propane and to a lesser degree for propylene. All propane concentrations are quite low except for the values very close to the combustion zone. This indicates that the propane concentrations are reduced by an order of magnitude only about two meters above the combustion zone.

Note that the plume generator tests indicate that PFTIR data for CO2 and propylene may

be biased low and high, respectively, so controlled flare data for these species may also be biased. The variation in the CO2 data was a concern, but most of it is attributable to variations in the position of the FOV, relative to the edge of the plume, and the amount of air being entrained by the plume. Intuitively, one would expect the CO2 to be lowest at the point where the temperature is the lowest (due to air entrainment), and this is the case in the third row of data. However, the value of 248,200 ppm obtained in the fourth row of data is problematic, because a propane flame should not produce CO2 concentrations so high. This requires further consideration.

The combustion efficiency for this flare was 99.8 ± 0.30 percent at the lowest measured

plume temperature, which is consistent with expectations for a well-designed and operated flare using propane as fuel. This efficiency, as with the plume generator calculated efficiencies, discussed earlier, is based on the efficiency equation shown in Section 1.0 of this report.

The hydrocarbon emission factor is based on the combustion efficiency and is the

fraction of entering hydrocarbon that exits unburned from the flare. For individual species, destruction efficiencies would be substituted for combustion efficiency and the emission factor would be for the individual species.

More details on the PFTIR data set for the controlled flare test data are presented in Appendix C. 5.4 Measurement Uncertainty and Quality Control

One of the goals of the test program was to quantify the uncertainty of PFTIR in measuring speciated emissions, using EFTIR as the reference measurement. Thus, it was necessary that all of the measurements performed be of the highest quality possible, and that the analytical error associated with the measurement device be understood. The uncertainty

5-34

associated with the EFTIR and PFTIR analyses, along with the quality control checks that were incorporated into the test program, are described in the following sections.

5.4.1 Measurement Uncertainty

The uncertainty associated with the measurement of speciated emissions from a flare by PFTIR will be quantified by comparison with the results of the EFTIR measurements. However, there is some uncertainty associated with the analyses used to develop results from both the EFTIR and the PFTIR. A discussion of this analytical uncertainty for the two methods is presented in the following sections. 5.4.1.1 EFTIR Analytical Uncertainty

For each target compound, there is an uncertainty associated with the concentration measured by the EFTIR. The analytical software uses a classical-least-squares fitting tool to identify the analytes by fitting all the absorption features of a particular frequency to those in a series of references. The analyte concentrations are determined by scaling the intensities of the absorption to the absorption of the references. In addition, the method also calculates a spectral residual, represented as a concentration, for each constituent analyzed. The spectral residual represents a 95 percent confidence level (2σ) for the quantified result and is solely dependent on the EFTIR instrument parameters (noise, spectral interference, resolution etc.). Thus, this spectral residual is used to define the analytical uncertainty of the measurement.

Table 5-14 presents the calculated overall uncertainties for each analyte of interest. During the plume generator testing, samples from five different test conditions were measured by the EFTIR. The spiked concentration of target analytes was varied between tests, as well as the temperature of the hot air in the stack. The uncertainties for the EFTIR data, with the exception of propane for Test 1b, were calculated from a data set comprised of three representative spectra from each test (30 total). Since this data set was a compilation of spectra from all the test conditions, it encompassed the concentration ranges observed for each analyte of interest. For each spectra, the residuals (in ppm) for each observed compound were then converted to a percentage of the observed analyte concentration. For example, if the method returned a result of 90 ppm for the analyte and ±9 ppm for the residual, the percentage would be 10 percent. Data were omitted from the calculations for tests in which a compound was not spiked. The average of these percentages throughout the data set was calculated and represents the overall analytical uncertainty associated with the EFTIR to quantify the particular analyte over the range of concentrations observed during testing. The same technique was applied to the data set from

5-35

August 27 for the single test in which propane was spiked into the heated air stream. This data set was comprised of 16 spectra in which propane was used in place of n-Butane. Data sets used for these analyses are provided in Appendix B.

Table 5-14. Instrument Uncertainties for Analytes of Interest

Analyte Uncertainty CO ±0.74% CO2 ±0.48%

n-Butane ±1.72% Ethylene ±0.23%

Propylene ±0.44% Propane ±0.20% Water ±0.77%

5.4.1.2 PFTIR Uncertainty

PFTIR measurements like those of the EFTIR can introduce error. Also like the EFTIR, the error is derived from the spectral residual of the analysis. If the analysis was perfect, the spectral residual in each analysis region would be zero. This is not the case, so the residual in the absorbance spectrum (or radiance spectrum) represents the possible error in the analysis. This is measurement as a standard deviation in each analysis region. This standard deviation is then converted to a gas concentration using the calibrated reference for each gas. The reported value is twice the standard deviation converted to concentrations.

To illustrate the relative magnitude of the measurement uncertainty for each compound,

Table 5-15 presents the CLS standard errors (2 sigma errors) and two times the standard deviation observed while looking at the hot air “blanks” during the tests at John Zink. The fact that the CLS and standard deviations are comparable to the noise limited laboratory values indicated that during this test the field system was noise limited not interference limited. It is also evidence that the field system had comparable or even lower noise level than that measured in the lab. This is possible because of a refocusing done in the field to improve data throughput after the first day of testing.

5-36

Table 5-15. Comparison of PFTIR Measurement Uncertainties to Field Blanks Blanks @ 498K (ppm*m) Peak Radiance Noise limited (ppm*m) CLS 2 Sigma 900-1000 cm-1 3000 cm-1

Conc. (ppm)

Path (m) 1000 cm-1 3000 cm-1 2 sigma error Deviation

225°C

CO 3.50E-07 20 1 0.971 0.755 0.252 (C2H4) 1.17E-06 9.9 1 0.113 0.348 0.158 (C3H6) 2.30E-06 75 1 0.434 0.756 1.056 (C3H8) 1.00E-08 1.20E-07 20 1 26.600 5.50 31.666 30.894 (C4H10) 2.30E-06 73 1 0.422 0.135 0.029 THC 4.08E-07 30 2 4.86 THC – InSb* 0.49

5.4.2 Quality Control The measurement data from this program were subject to quality control (QC) checks designed to measure the representativeness and comparability of the results. The results of these QC checks were evaluated and compared to the objectives established at the onset of this program. QC checks that fail to meet the objectives do not necessarily render the data unacceptable; however, they may affect the representativeness and comparability of the results. This section presents the QC results obtained, with the project objectives serving as benchmarks for comparison. In addition, this section provides a discussion of the process instrumentation calibration procedures used to verify proper operation of test equipment prior to testing. 5.4.2.1 Canister Samples

QA results for plume generator and propane fuel canister samples are presented in Tables 5-16 and 5-17, respectively. Detailed analytical results for canister samples are provided in Appendix F. For the plume generator, triplicate samples were collected over at 6-minute time period to assess overall sample variability. During the controlled flare test, two propane fuel canister samples were collected over a 20-minute internal.

Plume Generator Samples – Duplicate analysis of one sample was conducted for ethylene and propylene to assess analytical precision with the QC objective of <30 percent relative percent difference (RPD). QC results indicate that the data met this objective. Although there was no objective established for the overall sample variability, as measured by the percent RSD, all values were 3 percent or less indicating results for the three canister samples showed little variation.

5-37

Table 5-16. QC Results for Plume Generator Canister Samples

Parameter Method

Sample Mean

(ppmv) % CV

Analytical Precision (% RPD)

Precision Objective (%RPD)

UHP Nitrogen

Field Blank

(ppmv) Lab Blank

(ppmv)

Blank Objective (ppmv)

LCS (%

Recovery)

Objective (%

Recovery) CO ASTM-1945 35 3.3 - <30 ND (21) ND (10) ND 99 75 - 125 CO2 ASTM-1945 6600 1.5 - <30 25 ND (10) ND 98 75 – 125 Butane ASTM-1945 25 J 1.4 - <30 ND (21) ND (10) ND 97 75 – 125 Ethylene ATL @82 22 0.8 0.5 <30 ND (0.1) ND (0.5) ND 112 75 – 125 Propylene ATL @82 25 1.1 0.4 <30 ND (0.1) ND (0.5) ND 117 75 – 125 Oxygen ASTM-1945 18 vol% 3.1 - <30 ND (2100) ND (1000) ND 90 75 – 125 Nitrogen ASTM-1945 81 vol% 0.7 - <30 100 vol% ND (1000) NDa 100 75 - 125 a Objective does not apply to the UHP nitrogen field blank. UHP = Ultra high purity CV = Coefficient of variation. RPD = Relative percent difference. LCS = Laboratory control sample. ND = Not detected (reporting limit shown in parentheses). J = Indicates the sample concentration was between the reporting limit and the method detection limit (MDL).

5-38

Table 5-17. QC Results for Propane Fuel Canister Samples

Parameter Method

Sample Mean

(vol%) a % CV

Lab Blank

(vol %)

Blank Objective (ppmv)

LCS (% Recovery)

LCS Objective(% Recovery)

Methane ASTM-1945 0.04 NC ND (0.0001) ND 98 75 - 125 Ethane ASTM-1945 4.8 NC ND (0.001) ND 102 75 – 125 Propane ASTM-1945 92 NC ND (0.001) ND 96 75 – 125 Isobutane ASTM-1945 0.38 NC ND (0.001) ND 96 75 – 125 Butane ASTM-1945 0.22 NC ND (0.001) ND 97 75 – 125 Neopentane ASTM-1945 ND (0.0025) NC ND (0.001) ND 100 75 – 125 Isopentane ASTM-1945 ND (0.0025) NC ND (0.001) ND 99 75 - 125 Pentane ASTM-1945 ND (0.0025) NC ND (0.001) ND 97 75 - 125 C6+ HC ASTM-1945 ND (0.0025) NC ND (0.01) ND 94 75 - 125

a Based on a single sample result. One of the two samples was considered invalid. UHP = Ultra high purity CV = Coefficient of variation. RPD = Relative percent difference. LCS = Laboratory control sample. ND = Not detected (reporting limit shown in parentheses). NC = Not calculated. One of the two samples collected was considered invalid.

5-39

Potential field and laboratory sample contamination was assessed by the collection analysis of an ultra high purity (UHP) nitrogen field blank and of a laboratory blank, respectively. Field blank results for the UHP nitrogen sample and the lab blank show that none of the target compounds, with the exception of CO2, were detected. Carbon dioxide was detected at a concentration of 25 ppmv in the UHP nitrogen field blank (approximately 0.4 percent of the reported sample concentrations).

Accuracy of the analytical method was assessed using a laboratory control sample (LCS), with results expressed as a percent recovery of the target compounds. All results for the LCS sample were within the objective of 75 percent to 125 percent recovery established for the analytical methods by the laboratory.

Propane Fuel Canister Samples – One of the two canister samples (sample ID Test

6-1) collected on August 27 during the controlled flare test was considered invalid because of discrepancies between the final canister pressure measured in the field and upon receipt by the laboratory (10 in. Hg vacuum compared to 28 in. vacuum, respectively). Therefore, results for this sample were not used in any further data analysis.

Laboratory sample contamination was assessed by analysis of a laboratory blank. Results for the lab blank show that none of the target compounds were detected. Accuracy of the analytical method was assessed using an LCS, with results expressed as a percent recovery of the target compounds. All results for the LCS sample were within the objective of 75 to 125 percent recovery established for the analytical methods by the laboratory. 5.4.2.2 EFTIR Measurements

Prior to testing, the EFTIR instrument was challenged with certified gas standards to assess the accuracy and precision of the analytical method. This process was conducted in accordance with the procedures specified in EPA Method 320 for EFTIR. Results of the Method 320 spiking are summarized in Table 5-18. Detailed EFTIR results are presented in Appendix B.

5-40

Table 5-18 compares the average measured concentrations to the certified gas standard concentrations with accuracy expressed as percent recovery. In all cases, the measurements were within the acceptance criteria of 70 to 130 percent recovery established by EPA Method 320. Percent recoveries were in the range of 94 to 106 percent for ethylene, 96 to 101 percent for n-butane, and 109 to 113 percent for propylene. Table 5-18 also provides information on the repeatability of the multiple analyses conducted for each certified standard, expressed as percent coefficient of variation [CV = (standard deviation/sample mean) × 100]. For ethylene, CV values ranged from 0.5 to 2.6 percent. CV values for butane and propylene were similar, ranging from 0.6 to 1.9 percent and 0.03 to 1.9 percent, respectively.

In addition, native concentrations in the heated air carrier gas (i.e., without addition of spiking gases) were measured prior to each plume generator test condition. These results, summarized in Table 5-19, provide an indication of the native background concentrations of each compound in the heated air carrier gas stream. Native concentrations for CO2 averaged about 410 ppmv compared to sample results on the order of 6600 ppmv in the plume generator stack gas; similar native concentrations were observed on all three days. Concentrations of CO in the heated air carrier gas were on the order of 0.34 ppmv on August 22 and August 27, with a slightly higher mean concentration of 0.71 ppmv noted on August 26. These CO concentrations compare to mean EFTIR sample concentrations in the range of 20 to 140 ppmv. Native concentrations of the other target VOC compounds (propylene, ethylene, and butane) were generally below the method detection limits of the EFTIR instrument during all three days as reported in Table 5-19.

Additional instrument checks were conducted as part of the QA process to ensure proper operation of the EFTIR equipment. (Refer to Appendix B for a copy of the EFTIR logbook containing documentation of QA system checks.)

5-41

Table 5-18. EFTIR Method 320 Spiking Results for Certified Gas Standards

Date Parameter

Certified Standard +

2% (ppmv)

Average Measured

Concentration (ppmv)

Accuracy (% Recovery)

Method 320 Acceptance

Criteria (% Recovery)

Number of Measurements

Percent CV

128.8 127.2 98.7 70 – 130 13 2.0 150.3 147.9 98.4 70 – 130 2 2.2

n-Butane

50.1 49.2 98.1 70 – 130 3 0.6 30.9 29.8 96.4 70 - 130 6 2.6 11.8 11.3 95.6 70 – 130 4 0.5

Ethylene

34.2 34.4 100.8 70 – 130 5 1.5 34.6 38.6 111.6 70 – 130 3 0.8 15.1 17.0 112.9 70 – 130 2 1.7

8/22/03

Propylene

38.9 43.7 112.5 70 – 130 3 0.5 60.6 58.3 96.2 70 – 130 4 1.5 80.7 79.2 98.1 70 – 130 4 1.3

n-Butane

42.3 42.6 100.6 70 – 130 4 0.8 35.5 33.4 94.2 70 – 130 5 0.7 13.1 12.4 94.4 70 – 130 4 1.4

Ethylene

36.3 37.7 103.8 70 – 130 5 0.3 15.3 17.1 111.8 70 – 130 4 1.0 10.0 11.3 113.0 70 – 130 3 0.9

8/26/03

Propylene

19.9 22.3 112.5 70 – 130 4 0.7 211.2 208.0 98.5 70 – 130 5 1.0 237.9 231.4 97.3 70 – 130 5 0.6

n-Butane

78.6 76.9 97.9 70 – 130 5 1.9 44.1 46.6 105.6 70 – 130 4 0.7 20.9 20.4 97.7 70 – 130 3 0.7

Ethylene

59.6 57.8 97.1 70 – 130 4 1.0 48.2 53.1 110.2 70 – 130 3 0.03 20.9 23.4 111.8 70 – 130 3 1.4

8/27/03

Propylene

60.9 66.5 109.2 70 - 130 3 1.9

5-42

Table 5-19. EFTIR Heated Air Background Measurements

CO

(ppmv) CO2

(ppmv) Butane (ppmv)

H2O (%)

Ethylene (ppmv)

Propylene (ppmv)

Date/Time Period: 8/25/03 15:00:02 – 15:58:42 Mean 0.56 410 BDL(3.5) 2.2 BDL(0.07) BDL(0.8) Minimum 0.31 380 BDL(3.5) 1.9 BDL(0.07) BDL(0.8) Maximum 0.85 430 BDL(3.5) 2.4 BDL(0.07) BDL(0.8) % CV 26 2.9 NA 5.9 NA NA Date/Time Periods: 8/26/03 10:12:11 – 10:19:11, 10:40:08 – 10:44:20, 11:05:59 – 11:17:10, 13:25:08

– 13:31:26, 13:55:52 – 13:59:22, 14:28:00 – 14:34:17, 14:58:43 – 15:35:02, 15:59:29 – 16:03:40, 17:13:30 – 17:17:42

Mean 0.73 440 BDL(6) 2.2 BDL(0.3) BDL(2.6) Minimum 0.31 390 BDL(6) 1.9 BDL(0.3) BDL(2.6) Maximum 1.3 740 BDL(6) 2.6 BDL(0.3) BDL(2.6) % CV 32 12.4 NA 7.7 NA NA

Date/Time Period: 8/27/03 10:00:03 – 10:07:43 Average 0.34 410 BDL(1.7) a 2.4 BDL(0.3) BDL(2.6) Minimum 0.26 380 BDL(1.7) a 2.3 BDL(0.3) BDL(2.6) Maximum 0.43 420 BDL(1.7) a 2.6 BDL(0.3) BDL(2.6) % CV 10.8 3.5 NA a 3.1 NA NA

a Data are for propane which was substituted for butane during Test 1b conducted on August 7, 2003. BDL = Below detection limit (method detection limit shown in parentheses). CV = Coefficient of variation.

Cell Leak Checks – This test checks the integrity of each cell by pulling a vacuum and then monitoring the leak rate. The acceptance criterion for this test is a leak rate < 2 torr/minute. The extractive system was leak checked from probe tip to pump two times while on site, and no appreciable leak (as would be indicated by a change in pressure) was observed over a 180-second period.

Sample Cell Exchange Rate – Sampling flow rates through the cell were on the order of 7 standard liter per minute (slm) during the majority of monitoring, resulting in a complete sample exchange every 12-15 seconds. Since spectral averaging was conducted at 40-second intervals, each collected spectrum represented an integrated average over many sample cell exchanges. The flow was set and monitored using a rotameter while on site. It was checked at random intervals during testing. The sample flow was not changed for Method 320 spikes, and the dilution rates indicate flows greater than 7 slm.

5-43

Infrared Detector Linearity Checks – For best results, the infrared detector must yield a linear response throughout a reasonable absorbance range at all the frequencies of interest. A software linearizer is used to continuously adjust the MCT detector preamp signal in order to achieve the desired response. To optimize the linearizer, background spectra are acquired with and without a polyethylene film in the IR beam. Comparison of the strongly absorbing polyethylene bands in the low, mid, and high frequency regions against a clean background enables the processor to calculate and appropriately set the detector’s linearizer terms (offset, linear, quad, cubic and delay). This procedure was run prior to the start of testing, and subsequent spectra were visually checked on a periodic basis to confirm that linearity was maintained. This was performed prior to shipment of the instrument. The spectra were also checked as part of data analysis/reduction conducted on site. There were no indications of large IR absorbencies that would affect the detector linearity.

Noise Equivalent Absorbance (NEA) or Signal-to-Noise Ratio (SNR) Tests –

Signal-to-noise ratio (SNR) tests were performed by measuring the noise equivalent absorbance (NEA) of the EFTIR system. An NEA spectrum was obtained by collecting two consecutive single beam spectra, while the sample cell was being purged with UHP nitrogen, then forming a ratio with them to one another. This effectively produces a noise or “zero” spectrum in which the peak-to-peak absorbance is determined. This provides a measure of the sensitivity of the instrument for the specified spectral resolution (0.5 cm-1, in this case) and number of scans (537, or approximately 5 minutes of signal averaging). An SNR was run prior to shipment of the instruments, and then re-checked prior to the first test day. No appreciable difference in the pre- and post-shipment NEA was observed. The peak-to-peak absorbance is calculated in three different regions: 1000-1100cm-1, 2450-2550 cm-1 and 2900-3000 cm-1. Results from the site NEA test are more than adequate for an instrument sampling the expected concentrations for this test program, as shown below:

SNR Report, August 25, 2003 Resolution = 0.5 cm-1 Single Sided Number of Scans = 538 Range = 1000-1100 cm-1, RMS Noise = 0.0428%, SNR = 2336 Range = 2450-2550 cm-1, RMS Noise = 0.0750%, SNR = 1333 Range = 2900-3000 cm-1, RMS Noise = 0.268%, SNR = 373

5-44

Spectral Background – A spectral background is essentially a “blank spectrum” in that it does not contain any of the target compounds present in the sample. It was created by purging the cell with UHP nitrogen while collecting a spectrum. The analytical software then formed a ratio of this spectrum against each collected sample spectrum to produce an absorbance spectrum for quantitative analysis. Backgrounds were generated on a daily basis.

Spectrometer Frequency and Resolution Checks – A real-time check of

frequency position and resolution was performed each day by monitoring a specific water absorption line (present in ambient air). The position of this line must not deviate more than ±0.005 cm-1 from the reference value over the course of each test. Likewise, the line width (directly related to instrument resolution) must not deviate more than ±0.05 cm-1 from the reference value over the course of each test. The EFTIR instrument passed these criteria during each inspection. 5.4.2.3 PFTIR Measurements The PFTIR measurement data from this program were subject to quality control (QC) checks designed to ensure the quality of the results. A summary of these checks is presented in the following paragraphs. Documentation - For all PFTIR testing, a log was kept of all activities recording the beginning and ending times for each activity and noting any significant events occurring. Times for sample measurements, background measurements, and calibrations were noted. For each test the official start time was noted as was the time at which a stable PFTIR signal was recorded. This allowed for appropriate time-averaging of data for each test condition during data reduction. NEA Tests - Prior to shipping the PFTIR system to John Zink its signal to noise was measured in the laboratory as a noise equivalent radiance and a system response to a known black body signal. When the equipment was set up for the first day of testing, the signal to noise observed with the simulator as a source was lower than expected based upon lab measurements. As a result, the full optical system was checked and it was determined that the telescope focus was not optimum for the test conditions. This was consequently corrected and the resulting signal level increased by more than an order of magnitude. This provided a signal in agreement with what the laboratory tests suggested. Radiant signal level was then monitored on a daily

5-45

bases based upon both sky background signals and black body signals. No significant change in signal was observed throughout the rest of the test series. Black Body Calibrations - The PFTIR was calibrated on a daily basis using a black body source and a 12” telescope collimator. The black body source in the collimator was substituted with a quartz halogen lamp before each calibration and the visible beam used to align the PFTIR to the collimator. Once aligned, the black body source was again put in the collimator and calibrations were performed at 150°C, 200°C, and 250°C. These temperatures were used to encompass the range of source temperatures to be used in the measurements. Each black body calibration at each temperature was done with a minimum of 5 minute averaging to provide high signal to noise. The resulting spectra were then used to produce a daily spectral calibration curve for the instrument. The variations seen in the calibrations from day to day and their effect upon the data has been discussed earlier.

Sky Background - Sky background can have a significant influence on the observed infrared signal and it must be corrected for. As a result, sky backgrounds were collected before or during each measurement series. These spectra were collected by moving the PFTIR telescope upwind of the source while maintaining the same angle of view. Direct sighting through the telescope assured that the full field of view was open to the sky. The background signal from the sky was then averaged over five a minute period to provide a high signal to noise background spectrum. This spectrum could then be used to correct the measured plume data for inherent background emissions. If the sky conditions changed by having clouds move into the field of view, additional sky backgrounds were collected. This was necessary only on August 26 at about 15:00 hours. Each background spectrum was date and time stamped by the PFTIR computer so it could be correlated with the appropriate measurement series. Variations in sky background and their influence on the data were discussed elsewhere in this report.

Detector Linearity - The MCT detector used in the tests is inherently non-linear. However a linearity circuit is provided in the preamplifier so the response can be linearized over a specific operational range. To assure linear response, the linearity was checked during black body calibrations on a daily basis. The black body provided a strong signal, which would allow non-linear behavior to be detected. Linearity was checked by viewing the spectral range beyond detector cut off (300 to 500 cm-1) as well as checking the zero levels of totally absorbing bands for H2O and CO2. If the detector is operating non-linearly the signal in the low wavenumber range will not be zero as it should be and the zero levels in absorbing bands can be negative. If

5-46

this was observed preamplifier circuit was adjusted to correct it. The fist day of testing some adjustment of the linearity circuit was required. No subsequent adjustment was needed.

Real-Time Manual Data Checks - Throughout data collection, the PFTIR spectra were continuously viewed to make sure they appeared normal. The presence of most compounds used in testing could be seen visually in the spectra and their stability could be estimated by viewing series of spectra in time. The signal to noise of each spectrum could also be estimated from the raw data so the appropriateness of the particular averaging interval could be checked. The only time visual checks indicated poor data was on the first day of testing when the signal was low and in the flare testing when the FOV got too close to the flame itself.

5.4.2.4 Process Instrumentation

Process instrumentation for the plume generator and controlled flare systems, described previously in Section 3 of this report, were calibrated prior to testing using procedures consistent with such instrumentation in an industrial process setting. Documentation of calibrations conducted by John Zink personnel are provided in Appendix E. Gas cylinder certifications for the spiking gases used during the plume generator tests are also provided in Appendix E. Flow meters for air, spiked gas species, and propane fuel flow rates were measured by standard John Zink instruments associated with the respective test systems. These instruments included:

1. Pressure or differential pressure sensing with integral transmitter; 2. Temperature sensing devices; and 3. Flow elements (orifice, flow nozzle, or modified pitot tube).

Pressure Sensor and Differential Pressure Sensor – The primary instrument

used in calibrating a pressure sensor/transmitter (PT) or differential pressure sensor/transmitter (DPT) was a Fluke Model 743 Display Unit. The Model 743 is part of a system that includes plug-in modules of different pressure range capabilities. In checking the calibration or recalibrating a PT or DPT, a module with a pressure range that covered the appropriate calibration range was selected and inserted into the Model 743. These modules can be recalibrated by Fluke and provided with a certification of the recalibration. The certification is valid for one year. John Zink’s Fluke Model 743 system including modules was sent to the

5-47

Fluke Service Center in Carrolton, Texas, for recalibration and recertification. A copy of this recalibration and recertification is provided in Appendix E. Calibration of Air Flow Nozzle and Orifice Meters – Flow sensing/measurement for air flow into the air preheater/plume generator was accomplished with an ASME dimensioned flow nozzle. The air flow nozzle calibration was verified by checking it dimensionally with special emphasis on the throat diameter. The flow nozzle was inspected for dents or other evidence of damage that would impede air flow or disrupt air flow, and any damage that was found was repaired. The flow nozzle was checked to ensure it was properly mounted. The location of the downstream pressure tap was checked to verify that it was free of obstruction.

For orifice meters, the orifice diameter was measured and compared to the markings on the orifice plate. The orifice was inspected for damage to the orifice inlet or outlet edges. If damage was found, the damaged orifice was repaired or replaced. Any notations were recorded on the orifice plate. The line diameter of the orifice run is measured and recorded.

The flare air blower inlet pressure-sensing array was checked for damage or dirt build-up.

It was cleaned and repaired as necessary. Proper mounting and location was checked visually. 5.5 Statistical Analysis of Plume Generator Test Results

The following section discusses the analytical uncertainty (or precision) of the combustion efficiency produced by the PFTIR and EFTIR methods, and the bias in the PFTIR measurement of combustion efficiency. Precision pertains to random variability alone. Both random variability and bias affect accuracy. The results of the statistical analysis showed that the PFTIR demonstrated an average uncertainty for the calculated combustion efficiency of ±0.30 percent for the tests conducted on the plume generator. The statistical analysis also identified a bias of –0.70 percent for the calculated combustion efficiency.

5-48

5.5.1 Analytical Uncertainty of the Combustion Efficiency Results The variance of the combustion efficiency for both the PFTIR and EFTIR methods was

approximated using the following:

⎟⎟⎠

⎞⎜⎜⎝

⎛−+⎟

⎟⎠

⎞⎜⎜⎝

⎛=⎟⎟

⎠

⎞⎜⎜⎝

⎛

yxyxy

x yxyxyxVar

µµµµµµ ),cov(2)var()var(

22

2

where:

x = [CO2], y = [CO2]+[CO]+[Butane]+[Ethylene]+[Propylene], µx = mean of x, approximated by the measured value of x, µy = mean of y, approximated by the measured value of y, var(x) = analytical variance of x, var(y) = analytical variance of y, and cov(x,y) = covariance between x and y. If it is assumed that the analytical errors in the different constituents are uncorrelated, the

variance of y was calculated as the sum of the variances of the terms in y. The covariance between the numerator and denominator in the equation for combustion efficiency is the variance of the term they have in common, [CO2], (i.e., cov(x,y) = var(x) was used in the variance of combustion efficiency calculations).

The variance calculations for the PFTIR and EFTIR methods were computed for each test

condition, using all available data for each method (i.e., the PFTIR calculations include all data collected whereas in Section 5.5.2, only data that had a corresponding sample point with the EFTIR data could be included).

Suppose a given efficiency measurement can be expressed as follows: E(measured) = E(true) + e(analytical) + e(time)

where: E(measured) = measured efficiency value for a given time, E(true) = true value of the efficiency at that time, e(analytical) = analytical error, and e(time) = random variable representing the deviation of the true efficiency at this time

from the average over the experiment.

5-49

The analytical uncertainties (e(analytical)) are presented in Table 5-20 in standard deviation form for easier interpretation (the units are in % rather than % squared). The sample standard deviation of the actual combustion efficiency values (e(analytical) + e(time)) is also provided for comparison purposes.

Table 5-20. Analytical Uncertainties for Combustion Efficiencies

PFTIR EFTIR Test

Condition e(analytical) e(analytical) +

e(time) e(analytical) e(analytical) + e(time)1a 0.0054 0.300 0.0025 0.115 2a 0.0060 0.352 0.0011 0.037 2b 0.0021 0.118 0.0005 0.034 3a 0.0043 0.294 0.0007 0.037

3b-1 0.0017 0.108 0.0002 0.024 3b-2 0.0018 0.116 0.0002 0.030 3b-3 0.0011 0.062 0.0007 0.080 4a 0.0050 0.387 0.0009 0.037 4b 0.0104 0.822 0.0013 0.114 5a 0.0032 0.246 0.0005 0.033 5b 0.0041 0.526 0.0003 0.034

As expected, the data in the table demonstrate that the uncertainty associated with the EFTIR results is an order of magnitude lower than that for the PFTIR. This was expected since the EFTIR samples are collected in the stack, and analyzed at a constant temperature, whereas the PFTIR method is subject to atmospheric effects. Test 4b has an uncertainty of 0.82 percent, which suggests that the combustion efficiency for this condition cannot be known with a certainty greater than this value.

Recall that the strength of the spectoral response increases with temperature and concentration. This would suggest that there would be less uncertainty associated with measurements at higher temperatures and/or concentrations. It is apparent that the uncertainty is generally greater for the lower temperature tests (Tests 4 and 5) than for the higher temperature tests. However, the data do not support the conclusion that there should be more uncertainty for the lower concentration tests. 5.5.2 Test for Bias in the PFTIR Method

To determine the accuracy of the PFTIR method, a test was conducted to determine whether the PFTIR estimate is biased relative to the reference method (EFTIR). PFTIR values were paired with EFTIR values measured at or nearly at the same time. Since the measurements

5-50

were performed while the composition of the plume was held constant, it is not believed that a time gap of a few seconds will introduce an important difference.

Before discussing the test for bias, it is instructive to look at the data for one of the test

conditions (3a) a bit closer. The following graph displays the efficiency measurements over time for both methods. It is clear that the PFTIR measurements are consistently lower than the EFTIR measurements. This would not be surprising if the test for bias (at least for 3a) shows a significant negative bias in the PFTIR method.

As seen in Table 5-21, the analytical uncertainty for Test 3a is about 0.3 percent.

Inspection of Figure 5-13 suggests that the calculated combustion efficiency varies from approximately 97.1 percent to 98.1 percent. Thus, the variability of the PFTIR values exceeds the level that can be explained in terms of the analytical variances and the temporal variability of the efficiency.

An assumption of normality is made when testing for bias. Histograms based on 10-20

points have limited ability to characterize the shape of the distribution but are of some value here. Figure 5-14 displays the distribution of the paired differences for Test 3a. The distribution of the 15 differences does not indicate there is reason to be concerned about this assumption in our hypothesis test below. There is some evidence of asymmetry, although this is hard to assess on the basis of a histogram based on only 15 points. Moreover, we require only that the error in the mean be normal; by the central limit theorem, the mean more nearly has a normal distribution than do the individual data points. Further, it will be shown that the hypothesis tests produced results with high confidence levels in most cases; we do not believe that the conclusions are affected by the distributional assumptions.

5-51

Table 5-21. Pairwise Differences of Combustion Efficiency Measurements

Test Condition

Mean of Differences Units

# of Pairs

Std. Error of Differences t-statistic

P-value for Test

Test Conclusion

1a -1.7338 % 16 0.0778 -22.2717 0.0000 S 2a -0.7075 % 12 0.0897 -7.8876 0.0000 S 2b -0.2160 % 5 0.0541 -3.9932 0.0081 S 3a -1.3000 % 15 0.0761 -17.0839 0.0000 S

3b-1 -0.1775 % 6 0.0559 -3.1732 0.0124 S 3b-2 -0.4727 % 6 0.0569 -8.3077 0.0002 S 3b-3 -0.4744 % 5 0.0353 -13.4533 0.0001 S 4a -0.2415 % 13 0.1057 -2.2855 0.0206 S 4b -0.7133 % 6 0.3808 -1.8732 0.0600 NS 5a -0.8593 % 18 0.0608 -14.1402 0.0000 S 5b -0.7917 % 6 0.2272 -3.4839 0.0088 S

S = t-test statistically significant at the alpha = 0.05 significance level NS = t-test not statistically significant at the alpha = 0.05 significance level

PFTIR and EFTIR Efficiency Measurements for Test Condition 3a

96.00

96.50

97.00

97.50

98.00

98.50

99.00

10:50

:19

10:52

:00

10:53

:07

10:54

:48

10:56

:29

10:57

:36

10:59

:17

11:00

:24

Time

Effic

ienc

y (%

)

PFTIREFTIR

Figure 5-13. Efficiency Measurements for Test Condition 3a

5-52

Distribution of Paired Efficiency Differences for Test Condition 3a

0

1

2

3

4

5

6

7

(-1.7,

-1.55

]

(-1.55

,-1.4]

(-1.4,

-1.25

]

(-1.25

,-1.1]

(-1.1,

-0.95

]

(-0.95

,-0.8]

(-0.8,

-0.65

]

PFTIR-EFTIR (%)

Freq

uenc

y

Figure 5-14. Distribution of Paired Efficiency Differences for Test 3a

The pairwise differences between the PFTIR estimates of combustion efficiency and the

corresponding EFTIR efficiency values (in %) were calculated as follows:

iii EPD −=

where: Pi = the ith estimate of the combustion efficiency by the PFTIR method, and Ei = the ith estimate of the combustion efficiency by the EFTIR method. The index i varies from 1 to n, the number of pairs.

The test statistic for each test condition was calculated as:

DsDt =

where: D = mean of the D values, Ds = standard error of D ,

= n

sD , and

Ds = standard deviation of the D values.

5-53

The tests demonstrate that there is significant evidence of bias in the PFTIR measurements (test 4b is the only one that is not statistically significant at .05, but the p-value is .06). The p-value provides the confidence with which we can conclude that there is a bias. Consider, for example, the results for test 3b-1. The p-value is about 0.012. The likelihood that a mean difference as large as the one observed could have occurred by chance is only 0.012, or 1.2 percent. Thus, we conclude that a bias exists with 98.8 percent confidence. The confidence level is even higher for many of the tests. The mean differences for each test condition are negative, so it can be concluded that the PFTIR method produces a significantly negative bias. While this bias may be of statistical significance, it may not represent a practical engineering significance. That is, a negative bias in the PFTIR method may be acceptable, if the practical use for the method is to identify process flares that are operating at combustion efficiencies significantly lower (e.g., 90 percent) than the accepted standard.

Table 5-22 presents a summary of the combustion efficiency measurements from the

plume generator tests, along with the statistical data. The results suggest that there is both significant uncertainty and bias in the PFTIR measurements of combustion efficiency. However, the results appear to satisfy the desired data quality objective from the QAPP, which specified a maximum uncertainty of ± 1.0 percent. As discussed in a previous section, this may have resulted from offsetting errors in the concentration measurements of the individual compounds.

Table 5-22. Summary of Plume Generator Test Results

Test Condition

Combustion Efficiency

Percent Incomplete Combustion

Analytical

Uncertainty

Mean of Differences

(Bias)

Std. Error of Differences

1a 95.2 4.8 0.300 -1.734 0.0778 2a 96.4 3.6 0.352 -0.708 0.0897 2b 97.0 3.0 0.118 -0.216 0.0541 3a 97.5 2.5 0.294 -1.300 0.0761

3b-1 98.4 1.6 0.108 -0.178 0.0559 3b-2 98.2 1.8 0.116 -0.473 0.0569 3b-3 98.3 1.7 0.062 -0.474 0.0353 4a 97.0 3.0 0.387 -0.242 0.1057 4b 96.5 3.5 0.822 -0.7133 0.3808 5a 97.9 2.1 0.246 -0.8593 0.0608 5b 98.0 2.0 0.526 -0.7917 0.2272

5-54

The QAPP also established a data quality objective of ± 90 percent for the speciation of individual compounds in the total VOC. Due to the inability to measure total VOC due to a low signal strength in the H-C stretch of the spectrum, the data do not support the evaluation of this objective. However, inspection of the accuracy of the PFTIR concentration measurements for the individual hydrocarbon species, suggest that this objective would probably not be met with the current configuration, especially for butane and propylene. Improvements to the detector have already been identified that should produce a much higher signal strength, and dramatically improve the accuracy of the concentration measurements for both THC and the individual hydrocarbon species.

References

1. Rothman, L. S., et. al, "The HITRAN Molecular Spectroscopic Database and HAWKS (HITRAN Atmospheric Workstation): 1996 Edition", Journal of Quantitative Spectroscopy & Radiative Transfer, Vol. 60, No. 5, pp. 665-710, 1998.

6-1

6.0 Summary, Conclusions, and Recommendations

TCEQ has contracted the UH and URS to evaluate the feasibility of Passive Fourier Transform Infrared (PFTIR) spectroscopy as a method for measuring flare emissions and calculating destruction efficiency of VOCs. URS teamed with IMACC and John Zink to develop a multi-phase study for the purpose of determining whether PFTIR is a viable method for measuring mass emissions of individual species and total hydrocarbon emissions from process flares. This section summarizes the results of the Phase I analytical study and field testing, provides conclusions that describe the source of the uncertainty in the PFTIR measurements, and presents recommendations for the next phase of the program. 6.1 Summary

The Phase I test program was comprised of three major elements. The first of these was an analytical study of the PFTIR to determine the expected signal strength, minimum detection limits, and the sensitivity of the signal to plume gradients. The second element was a field test in which a simulated flare plume was generated containing known concentrations of selected gases. These gases were emitted from a stack from which a sample was extracted and measured using an EPA-approved reference method. The PFTIR then measured the concentrations of these gases immediately above the stack, and the results were compared with the reference method data. In the third element of Phase I, the PFTIR was used to measure emissions from an actual flare at a test facility, to develop information on the logistics of making the measurement.

A QAPP was developed to ensure that the PFTIR method was evaluated systematically,

and that the data would be of the highest quality. The QAPP specified the elements of the Phase I test program and the data quality objectives. It also detailed the sampling and analytical procedures, and the statistical approach that would be used to report the results.

The analytical study was completed in August 2003. The study produced data for the

peak radiance and minimum detectable gas concentrations for the gas species that were part of the field test program. These results were used to determine that the noise-limited calcuable combustion efficiency was about 99.95 percent. The study also suggested that the Mercury Cadmium Telluride (MCT) detector that would be used for the field tests would have relatively low signal strength for propane (the fuel for the actual flare test) and total hydrocarbons (THC). The conclusion from the study was that future phases of the PFTIR flare test program should incorporate an Indium Antimonide (InSb) detector, in tandem with the MCT detector. This will

6-2

provide the best possible results for the range of gas species that will need to be measured to provide quality flare emission data. Unfortunately, due to the compressed project schedule, and the lead-time for a new detector, it was not possible to implement this change in time for the field tests on the plume generator and controlled flare.

The field test program was conducted August 25-27, 2003 at the John Zink test facility in

Tulsa, OK. In the plume generator tests, emissions measurements from the PFTIR were compared to measurements from an extractive FTIR (EFTIR) using EPA Method 320. The concentration data from these measurements were then used to calculate the combustion efficiency for the simulated flare. A statistical analysis was performed for the paired (PFTIR and EFTIR) combustion efficiency data which produced an indication of the bias and the randomness of the PFTIR data, as compared to the EFTIR results.

The PFTIR testing on the actual flare was performed on August 27, 2003. These tests

were conducted with the flare firing commercial grade propane at a rate of 10,000 lb/hr. A series of measurements were completed in which the field of view for the PFTIR was moved to a number of locations in the flare plume. The indicated values for temperature, carbon monoxide, and propane decreased with increasing distance from the flame, as would be expected. The average combustion efficiency for the flare for these tests varied from 99.5 to 99.9 percent (Note: the uncertainty for the measurement of the controlled flare was not determined.)

The temperature of the flare plume was found to be higher than that used in the plume

generator testing. This is encouraging for real-world flare applications, because the signal strength increases with increasing temperature, and so the measurements on actual flares should have less uncertainty than was observed during the plume generator tests.

Another outcome of the test effort was that two analytical methodologies were identified

for reducing the PFTIR results, the Classic Least Squares (CLS) approach, and the line-by-line (LBL) analysis of the spectra. Of the two, LBL should be the more rigorous because of its ability to treat gradients, dynamically fit resolution and line shift, and treat artifacts like calibration offset. For this reason, the LBL results are the dominant focus in this report. However, in many cases the CLS results actually fit the EFTIR data more closely than the LBL results. This is not completely understood, although some contributing factors are known. Probably the most significant effect is that CLS analyses are very fast. This makes it easy to adjust and test analysis regions and even reference spectra sets to optimize detection and

6-3

minimize errors for each individual specie. During the reduction of data with the CLS algorithm, adjustments were made of the analysis regions and of the inclusion/exclusion windows within regions. This was done using observed spectra gathered at various temperatures and efficiencies. In effect, a set of analysis regions optimized for all measurement scenarios was developed and this was used to reduce the full data set. This type of optimization is done in initially setting up analysis with LBL technique, but repeated alteration of it is difficult. Ultimately, the LBL approach should be the most accurate reduction algorithm particularly when gradients are significant. As a result, the LBL approach has been focused on here. CLS results have been included in the Appendix. 6.2 Conclusions

The analytical study and field test program results demonstrated that the PFTIR is a potentially viable technology for determining selected species concentrations and combustion efficiency in actual flare systems.

The analytical study evaluated the noise limited detection limit of several VOCs of

interest, over a range of temperatures (Table 5-2). The results show that at these detection limits, the maximum detectable combustion efficiency would be 99.95 percent. The study also revealed that the detection limits would be substantially improved if an IbSb/HgCdTe sandwich detector were used, rather than the standard HgCdTe detector used during the testing. (Due to the compressed schedule, it was not possible to order a new detector prior to the start of testing.)

The results of the plume generator tests, in which the PFTIR was compared to the EFTIR

reference method, showed that the PFTIR achieved the data quality objective of ±1.0 percent for calculating combustion efficiency (see Table 5-9). The average calculated uncertainty from the plume generator tests was 0.303 percent, with a maximum of 0.822 percent (Table 5-22). There was a significant average bias of –0.699 percent in the combustion efficiency determined from the PFTIR data (from data in Table 5-22). In addition, the results of the individual species concentration measurements demonstrated that there was significant uncertainty present, and the PFTIR was not able to achieve the desired level of accuracy of ±50 percent for butane and propylene (Table 5-7).

The controlled flare testing demonstrated that the PFTIR could measure low

concentrations of a variety of species under real-world conditions on an actual flare. The tests allowed for the determination of sensitivities of the measurements to specific measurement

6-4

parameters. The variability of the results made it clear that two issues were critical to proper quantification of the spectra: 1) the sky background used in reducing the data and 2) the absolute radiance calibration of the instrument.

Each of these had a different influence on the resultant gas concentrations. This is

described further in the following paragraphs:

• The sky radiance is dominant at longer wavelengths (700 to 1700 cm-1) because of its low temperature and the transparency of the plume at these infrared wavelengths. Because ethylene, propylene, and possibly propane concentrations are measured in this region, sky radiance influences the analysis of these gases. However, CO and CO2 signatures which are found between 2000 and 2500 cm-1, and total hydrocarbon measurements which may be monitored between 2900 and 3000 cm-1, will have little, if any, influence from sky radiance.

• The absolute radiance calibration of the PFTIR influences the absolute gas concentrations extracted from the data but to first order does not affect their relative values. The variance in this calibration was more a multiplicative factor, which raises or lowers the whole calibration curve influencing all gases similarly. In terms of combustion efficiency, this means that the absolute radiance calibration will have only a secondary effect on efficiency calculations because all gases vary the same way with calibration variations. The ratios taken in computing efficiency will make this calibration factor cancel to the extent it is equal for all species; and,

• The sky radiance will have a more significant influence on calculated efficiency because it influences certain organics and not others.

Another critical issue that surfaced in the tests was detection and calibration of total

hydrocarbon (THCs) concentrations. This was hindered in the present test series because of the low signal observed in the 3000 cm-1 region where the strongest bands of the THCs are found. The signal was low here because the temperature of the simulator was low and because the detector used in the PFTIR peaked at the longer wavelengths important to ethylene and propylene, which were considered high priority in the test plan. For most of the data generated in this report, the weaker bands around 1000 cm-1 were used for the THCs. These are weak enough that the concentrations extracted from the data had large error bars and high variances, which indicated that the uncertainty of the THC measurements was greater than the “true” value.

The detection of THC and VOC species in an actual process flare will be easier than in the plume generator because the temperature is higher. This augments the signature of the organics region around 3000 cm-1, making this region more accessible. In addition, there are

6-5

commercial detectors available for this spectral region that use Indium Antimonide (InSb) or Indium Arsenide (InAs) that are more than 100-times more sensitive than the mercury-cadmium-telluride (MCT) detector used in this study. So-called “sandwich” detectors are available which have a MCT and an InSb detector combined. Such a detector, if used in flare monitoring, would increase sensitivity by about 100 in the region where the signal is falling by about the same magnitude. This should allow THCs in the C-H stretch region, along with certain VOCs, to be observed with good sensitivity because of their significantly stronger signatures in that region, as compared to that used in the Phase I test program.

Another outcome of the test effort was that two analytical methodologies were identified for reducing the PFTIR results:

• Line-by-line calculations (LBL) taking into account all gradients and temperature

variations, and

• Classical Least Squares (CLS) analysis, which was somewhat more restricted because it only considers a single isothermal, isobaric layer in the plume.

The intent was that the CLS analysis, which is very fast (100 milliseconds), would be

used for real-time analysis of data during data collection in the field while the LBL analyses, which are more time consuming, would be used for full data reduction after the test. However, in looking at the results of these two analyses, the CLS actually returned gas concentrations and efficiencies closer to the EFTIR results than the LBL analyses (see Appendix C, Figures C-1 through C-30 and compare to Table 5-7 and Figures 5-7 through 5-10). This is probably fortuitous since the LBL analysis actually accounts for more of the actual radiative transfer effects in the plume than the CLS can. However, it is expected that, to the extent that temperature and concentration gradients are not significant in the plume, the two approaches should converge to the same results. This may well be the case in actual flare plumes since what was observed in the flare tests at Zink was a more random distribution of temperature and concentration with no clearly observable gradients. If this is actually the case in industrial flares, the two analysis methodologies should give comparable results.

It is important to understand that development of analysis procedures for quantification of

FTIR spectra must follow a two or three stage development and validation process. This type of validation was not possible for the emission codes used in this program because of a lack of appropriate test spectra. Development of a typical analysis method begins by specifying for the

6-6

computer what gases are to be analyzed, what interferants are expected to be present, and what spectral regions are to be used for analysis of each compound. At this stage, the analysis regions are typically chosen using library reference spectra for all compounds.

In standard absorption spectroscopy, this initial method is then validated using available spectral data appropriate for the type of source to be monitored. If such data are not available, spectra are usually generated with calibrated spike gases. The test spectra then allow the analysis method to be tested optimizing its accuracy in the complex spectra representative of the source. This optimization is performed by: 1) adjusting analysis regions (and in some software by defining inclusion or exclusion subregions) to minimize interference, 2) selecting the best possible references for all compounds matching the characteristics of the instrument, and if necessary correcting for non-linearity in the system response.

These procedures could not be performed for the LBL or CLS reduction procedures in the

present program because appropriate radiance data for flares was not available. In effect, the data collected on the plume simulator was the first data available for this type of optimization and validation. It is for this reason that the simulator tests were seen as a optimization and validation process in its own right. Of the two procedures used for analysis of field data, the CLS method was rapid enough that iterative adjustment of regions, references, etc. could be done in an effort to minimize reported error and maximize accuracy. It is for this reason that the later CLS results gave better and better comparisons with the EFTIR results. The LBL procedures underwent only a very limited optimization because of the much longer cycle time for it analysis. It may be that a more comprehensive optimization of the LBL procedures will be necessary using simulator data for which true gas concentrations are known.

Finally, the results from the test program suggest that combustion efficiency (as defined

by the equation in Section 1.0) may not be a suitable metric for use in determining the effectiveness of any flare emission measurement technology. This is because the CO2 concentration dominates the equation, such that significant uncertainties in the individual species concentrations may produce very small uncertainty in the combustion efficiency. Compensating biases in the measurement of the individual species may also produce a small bias for the combustion efficiency. As a result, future programs should focus on the accuracy of individual species concentrations and the corresponding destruction efficiency.

6-7

6.3 Recommendations

The tests at John Zink showed the basic capability of passive FTIR to detect and quantify gas concentrations from a flare. The three issues that need to be addressed are:

• The absolute calibrations of the instrument which will bring the plume gas

concentrations into line with what are believed to be true values;

• Frequent or simultaneous sky radiance measurements allowing for accurate “nulling” of sky radiance in the analyzed spectra; and,

• Detection of THC and speciated VOC compounds.

All of these issues are manageable. The absolute calibrations require a more careful

alignment of the black body source at short range to assure a full filling of the input aperture of the PFTIR telescope. The second issue of sky radiance was not expected. This will necessitate regular measurements of the sky background throughout all flare measurement series. This is not difficult, as it only requires that the PFTIR telescope be rotated off the source to get a “current” sky signature at regular intervals throughout testing. The detection of THC can be improved by adding using a “sandwich” detector that would produce stronger signatures in the C-H stretch region of the spectrum. These changes should allow reliable and accurate plume data to be collected with passive FTIR.

To demonstrate the validity of the PFTIR, a second series of tests should be conducted on the John Zink plume generator. These tests should be conducted at higher temperatures that are more typical of actual process flare plumes. The objectives of the test would be to evaluate the recommended modifications to the PFTIR detector and calibration procedures, and confirm that the PFTIR is capable of accurately measuring both individual VOCs and THCs, when compared to a reference method, such as EFTIR.

A series of tests should then be conducted with a controlled flare or a well-documented

flare. There is a ground-level flare at the John Zink test facility that would be ideal. The flare should have adequate instrumentation to quantify the combustion efficiency or equivalently the CO, CO2, and THC. If possible the combustion efficiency results should be augmented with an analysis of individual VOCs like ethylene, propylene, propane, etc. Such a test series would allow for demonstration of the PFTIR to accurately quantify high temperature, complex gas composition flares in real environments.

6-8

Another objective of the Phase II testing will be to determine the effect of non-optimal

process operating conditions on flare emissions. These non-optimal conditions would include the performance of a flare at lower firing conditions in the presence of high steam- or air-assist flow rates.

These tests should then be followed by field measurements at appropriate industrial

facilities representative of those in current operation in Texas.

Appendix A QAPP

(Included on CD)

Appendix B Extractive FTIR Data

Table B-1 Table B-2 Table B-3

(Included on CD)

Appendix C Passive FTIR Data and Calculations

(Included on CD)

Appendix D John Zink Process Data

Table D-1 Table D-2 Table D-3 Table D-4

(Included on CD)

Appendix E Sampling Equipment Calibration Data

(Included on CD)

Appendix F Canister Analyses

(Included on CD)

Documents

Passive FTIR Phase I Testing of Simulated and …...Passive FTIR Phase I Testing of Simulated and Controlled Flare Systems FINAL REPORT Prepared for: Texas Commission on Environmental